N N + = N +, can be calculated from the radar s matched

Transcription

1 88 Ofcom Spectral Efficiency Scheme (SES ) 2 PGtGrλ ( 4πR) L L t S = 2 atmos diffraction Equation Where: S = Incident signal power P t = Transmit Power G t = Transmit antenna gain G r = Receive antenna gain λ = Wavelength R = Range between radar L atmos = Atmospheric losses L diffraction = Diffraction losses (curved Earth) Since the radar are ground based, at ranges beyond the radar horizon (in this case ~13km) curved Earth diffraction losses become significant, although this does not take into account any losses due to terrain screening. In addition, since each radar will not be synchronised to any other and the incidence of main beam to main beam interference is very small, it has been assumed that the practical worst case interference is from the main beam of one radar into the azimuth sidelobes of the other. Under these conditions G t = ~35dB and G r = ~0dB. The Protection Level, ( N 6dB) filter noise level : N N + = N +, can be calculated from the radar s matched 4 N = k. T 0. BW. NF Equation Where: k = Boltzman constant T 0 = System temperature BW = System matched filter bandwidth NF = Receiver Noise figure Figure plots the received signal power, S, vs. range between radar for a Watchman TWT 0.4µs pulse. The red line indicates the protection level of a Watchman. Out to approximately 80km separation, the received signal exceeds the protection level and the neighbouring radar will interfere with each other if they operate on the same centre frequency. The lower panel indicates the difference between the received signal and the protection level. This can be interpreted as the amount of extra isolation required between the two radar in order to reduce the received signal to below the protection level. This extra isolation may be achieved by ensuring the radar use different centre frequencies. The spacing between these centre frequencies must be sufficiently large that the waveform spectral envelope has rolled off to below the required isolation level, thus ensuring that the combination of range and frequency isolation reduces the received signal below the protection level. This analysis allows the spectral spacing to be related to the range spacing and the spectral envelope of a particular waveform, this will vary from radar to radar and waveform to

2 Ofcom Spectral Efficiency Scheme (SES ) 89 waveform. The relationship for the Watchman 0.4µs short range pulse is illustrated in Figure Signal level, db Peak signal at Radar due to interference from neighbouring Radar vs. range between radar Interfering signal Protection level Required isolation Range, km Reqd isolation, db Range, km Figure : Upper panel - Peak signal level from a neighbouring radar (Black line) and Protection level (red line). Lower panel - Additional required isolation to reduce signal below the protection level

3 90 Ofcom Spectral Efficiency Scheme (SES ) 50 Required frequency separation vs. Range Separation, MHz Range, km Figure : Required frequency separation vs. Range for the Watchman 0.4µs pulse It is clear from Figure that as the range separation decreases, the required spectral separation increases, the shortest range separation, however, is defined by the radar receiver maximum power limit level, which occurs at around 39km radar separation. This process can be repeated for the current and proposed Watchman waveform designs. These are plotted together in Figure to assess the potential spectral efficiency improvements.

4 Ofcom Spectral Efficiency Scheme (SES ) Required frequency separation vs. Range Watchman 0.4us plain pulse Watchman 1.0us plain pulse Watchman 20us 2.5MHz NLFM Watchman 20us 1.0MHz NLFM 35 Separation, MHz Range, km Figure : Required frequency separation vs. Range separation for current and proposed Watchman waveforms. It is interesting to note that at the longer ranges the required spectral separations converge towards a similar low value, this is as a result of the waveform spectral envelopes of all the waveforms flattening out away from the centre frequency and the effect of range and earth curvature losses. Also of note is that the differences between the curves for the 2.5MHz and 1.0MHz 20µs waveforms are very small. The differences between the 0.4µs and 1µs curves, however, are more obvious. For illustration of the improvements a comparison of the required spectral separations at 45km are tabulated in Table Pulse width (µs) Table : Required spectral separations for 0.4µs and 1.0µs Watchman pulses Separation 45km Difference 45km w.r.t. 0.4µs pulse Mean difference (MHz) over all ranges Using the 1µs pulse in preference to the 0.4µs pulse allows neighbouring radar at 45km range separation to be spectrally 0.7MHz closer. Alternatively, for a frequency separation of 20MHz, using the 1.0µs pulse allows interfering radar to be 3km closer (at 41km separation rather than the 44km separation allowed by the 0.4µs pulse) The average frequency separation improvement of using a 1µs pulse compared to a 0.4µs pulse is ~1.2MHz. This indicates that over a reasonable range of range separations the 1µs

5 92 Ofcom Spectral Efficiency Scheme (SES ) pulse waveform allows the radar to be on average spectrally closer to each other by ~1.2MHz. If this modest improvement were rolled out over the network of around 80 installed Watchman radar and the frequency allocations re-planned on this basis, this could lead to useful spectral efficiency improvements. However, degrading the range resolution in this way may not be consistent with military requirements and thereby not be applicable to a large number of the deployed systems. Cost of upgrade to Watchman In order to achieve these longer and narrower bandwidth pulses, it is necessary to make a number of changes to the radar system, in particular the waveform generator and receiver sub-systems as well as the signal processor. Such changes require significant effort, but do not necessarily result in significant life extension unless they are combined with upgrades to the rest of the system. It would thus not be economic to upgrade these systems in this way unless upgrades to give a working life extension were also incorporated. The economic cost of upgrading a Watchman radar system, indicated in section 4.2 has thus been based on calculations incorporating both the spectral efficiency changes outlined here and other upgrades necessary to ensure significant working life extension Migrate from TWT to Solid state transmitter technology As discussed in section TWT transmitters are limited in their maximum duty cycle (~2%) and pulse rise and fall times (~35ns). Which limits the achievable spectral efficiency of their transmissions. Solid state transmitters, however, can support much higher duty cycles (~10%) and inherently longer rise times (~170ns). This section examines the impact these parameters have on spectral efficiency and the improvements which may be realised through migrating/upgrading from TWT to Solid State transmitter technology. Table tabulates example waveform parameters for typical TWT and Solid State (SS) based radar systems. Examples 1-4 have already been analysed in the previous section ( ), the analysis here follows the same process. Table : Example transmitted waveform parameters for TWT and SS based radar system Num Description Tx τ u (µs) t r &t f (ns) P t (kw) Modulation Modulation BW (MHz) 1 TWT 0.4µs Plain TWT Plain N/A 2 TWT 1.0µs Plain TWT Plain N/A 3 TWT 20µs NLFM 2.5MHz TWT NLFM TWT 20µs NLFM 1.0MHz TWT NLFM SS 1.0µs Plain SS Plain N/A 6 SS 20µs NLFM 1.0MHz SS NLFM SS 50µs NLFM 1.0MHz SS NLFM SS 80µs NLFM 1.0MHz SS NLFM SS 100µs NLFM 1.0MHz SS NLFM 1.0

6 Ofcom Spectral Efficiency Scheme (SES ) TWT 0.4us Plain TWT 1.0us Plain TWT 20us NLFM 2.5MHz TWT 20us NLFM 1.0MHz SS 1.0us Plain SS 20us NLFM 1.0MHz SS 50us NLFM 1.0MHz SS 80us NLFM 1.0MHz SS 100us NLFM 1.0MHz Frequency offset, MHz Figure : One sided theoretical spectra for various transmitted waveforms Relative amplitude, db

7 94 Ofcom Spectral Efficiency Scheme (SES ) Required frequency separation vs. Range Watchman 0.4us plain pulse Watchman 1.0us plain pulse Watchman 20us 2.5MHz NLFM Watchman 20us 1.0MHz NLFM SS 1.0us plain pulse SS 20us 1.0MHz NLFM SS 50us 1.0MHz NLFM SS 80us 1.0MHz NLFM SS 100us 1.0MHz NLFM Separation, MHz Range, km Figure : Required frequency separation vs radar separation for various waveforms and transmitter types

8 Ofcom Spectral Efficiency Scheme (SES ) 95 Figure and Figure illustrate the theoretical transmitted spectra and consequent required radar range separations for the various waveforms detailed in Table The dotted lines represent those waveforms already analysed in section It is clear that the spectral efficiency of the Solid State waveforms is significantly improved over the TWT waveforms. This is as a result of the longer pulse lengths, lower peak power and slower inherent rise and fall times of the solid state power devices. If Figure is examined in more detail it becomes apparent that in terms of theoretical spectrum, the magnitude of the far out skirts of the spectra are sensitive to the rise and fall times and are considerably reduced by the slower solid state device responses. This effect is particularly evident when the TWT and SS 1.0µs plain pulses are compared. Table compares the required spectral separations at 45km range separation. Rise time (ns) Table : Required spectral separations for 1.0µs TWT and Solid State pulses Separation 45km Difference 45km w.r.t. TWT 35ns rise&fall time pulse Mean difference (MHz) over all ranges TWT = SS = A reduction in spectral separation from 18.5MHz, required for a TWT system, down to 6.7MHz for a slower rise time solid state system is indicated for a range separation of 45km, an improvement of 11.8MHz. On average, over all ranges, the improvement is 6MHz. In terms of required frequency and range separation, all of the longer solid state pulses have similar good performance as a result of their slow rise times and low peak powers. For example, at a range separation of 45km Figure reveals that neighbouring SS long pulse NLFM radar systems require at least 10MHz less centre frequency separation than TWT systems; at 60km range separation the improvement is reduced to 2MHz. These would be significant spectral efficiency improvements, particularly compared to those achievable by simply reducing the Watchman chirp bandwidth. The longer pulse lengths are necessary with the lower solid state peak powers to ensure enough energy is incident on wanted targets to allow sufficient detection performance at long ranges, for typical solid state peak powers, the longer pulses (50 100µs) are usually sufficient for long range detection. It is also interesting to note that, over a range of radar separations, the 1.0µs solid state plain pulse requires less frequency separation than even the TWT 20µs pulses, this is a result of the lower solid state peak power and longer rise&fall times, however, the detection performance of such a short, low peak power pulse is significantly reduced even compared to the 1.0µs TWT pulse. In conclusion, the move towards solid state transmitters utilising long modulated pulses in ATC radar is consistent with much improved spectral efficiency as a result of the slower inherent rise times and reduced peak powers of these devices. NATS is currently in the process of replacing its deployed radar with modern solid state radar systems, which have these spectral efficiency benefits, however, there are a number of radar deployed in regional airports and MoD sites that still use TWT transmitters.

9 96 Ofcom Spectral Efficiency Scheme (SES ) If further spectrum efficiency, beyond that achievable using current solid state long pulse transmitter technology needs to be realised, a change in transmitter technology would be required since the parameters which have the most effect on transmitted spectrum are the rise and fall times of the transmitted pulse edges, which for current solid state technology is limited to around ns maximum New transmitter techniques and technologies for new radar systems This section investigates new techniques and technologies which could be applied to new radar designs for the next generation of radar systems or upgrades to existing systems. In particular, attention is paid to methods of improving the ability of high power transmitters to allow slower rise and fall times or pulse shaping / modulation in order to reduce transmitted spectral width. The drawback of this approach is that less energy thus is available to illuminate the targets and detection performance will reduce. Since transmitter devices are typically peak power limited, the only way to reclaim any lost detection performance is to use longer pulses, which will have an impact on short range cover and may exceed the maximum duty of the transmitter devices Power supply modulation It is possible to modulate transmitter DC power supplies before and during RF pulses as they pass through the radar transmitter, in order to shape the RF pulse edges to give a more spectrally efficient pulse. However, this technique is difficult to implement and very wasteful in power Linearization Linearization is a technique that is very common in mobile communication systems. The idea is to increase the range over which a power amplifier remains linear in order to maximise the power amplifiers output power whilst it remains linear. However all linearization techniques are limited in their maximum correctable range, which is the region of power output near the onset of saturation. Unlike in radar, high spectral efficiency has been a goal in communication systems for many years in order to squeeze as much data into a particular bandwidth. Here the non-constant envelope (high peak-to average ratio) digital modulation schemes used in many 2.5G and 3G wireless systems make RF power amplifier linearity and efficiency a crucial design issue. Spectrally efficient radar waveforms could also have signal envelopes with high peak-to-average ratios and therefore to be effective require linear amplifiers. A typical class A amplifier stage can be run backed off i.e. with lower input drive to improve linearity however considering the large number of solid-state devices required in a radar this would create a very power inefficient and expensive radar. As already discussed; most radar transmitters are class C devices. Class C devices are designed to be either on or off and hence maximise efficiency when used with rectangular radar pulses. High power vacuum tubes are designed for this mode of operation and it would be very difficult to change them. Solid-state devices used in radar are also designed for this mode of operation, however there are plenty of solid-state devices for the communication industry that have been design for use under other amplifier configurations.

10 Ofcom Spectral Efficiency Scheme (SES ) 97 These could be implemented with a pre-distortion technique to generate a linear amplifier for a radar transmitter, however power efficiency would suffer. Some simple linearization schemes do exist; the most obvious is to increase the bias levels of the amplifier. This is equivalent to reducing the input signal level from a distortion point of view, however the disadvantage is increased DC power. A similar effect can be obtained through backing off the input signal. At lower frequencies, amplifiers are commonly run with local feedback for linearization on each gain stage. An example of this is shown in Figure a) below. Unfortunately radars operate in higher frequency bands where the gain from a single amplifier stage is much less than at lower frequencies. The low gain from each stage makes local feedback impractical and would severely deteriorate the overall gain of the transmitter. Even if several stages are cascaded with feedback around all of them, many gain stages would be required making the radar transmitter very expensive. Each of the gain stages would add delay to the signal that would cause the feedback loop to become unstable when used with amplifiers with low gain. Envelope detector Bias circuit RF in + - PA RF out RF in PA RF out (a) (b) Figure : a) Local feedback for PA linearization, b) open loop dynamic bias implementation for PA linearization. Figure b) shows a dynamic bias configuration. Dynamic bias adjusts the amplifiers bias supply depending on the input signal level. This can increase the amplifiers linear range without significantly increasing the DC power required. It is possible to get several db s of improvement in a linear amplifiers 1dB compression region with this technique. Feedback from the PA s output can also be incorporated to control the bias circuitry, however phase distortion can be a problem if a large phase shift occur in the amplifier causing the rising and falling edges of the radar pulse to distort. Base-band feedback where only the base-band signal is fed back through the feedback loop can be used to reduce the bandwidth required in the feedback loop. The demodulator used in the feedback loop and modulator before the power amplifier are assumed to be completely linear compared to the power amplifier. This system does suffer from a narrow bandwidth and therefore may not be suitable for radar because they have quite wide bands needed for frequency diversity Pre-distortion Pre-distortion exists in both open loop and closed loop forms. Open loop pre-distortion uses amplitude and phase pre-distortion to either pre-distort the RF or base-band signal such that it directly cancels the power amplifiers distortion. Open loop pre-distortion is relatively immune to stability problems even over a wide band and correct modest amounts of distortion. It can be combined with other linearization methods to obtain higher efficiency and linearity than with a single method of linearization. Closed loop pre-distortion

11 98 Ofcom Spectral Efficiency Scheme (SES ) essentially adapts the open loop pre-distortion technique to account for amplifier variation over time. RF in RF RF out RF in RF out Predistortion Predistortion PA PA (a) Adaptive predistortion processing and control Figure : a) Open-loop RF pre-distortion before PA, b) Closed-loop pre-distortion scheme. The adaptive feed-forward technique is one of the most common linearization methods used with communication systems. This provides good linearity however PA efficiency suffers. An example of an adaptive feed forward system is shown in Figure below. Here the RF input signal is split into two parts. One part passes through the PA and is amplified as normal. The other goes into canceller 1, which is a variable phase and gain shifter. The system tweaks canceller 1 settings to minimise the power at point A, leaving only the PA distortion remaining at point A. Canceller 2 then works on this distortion to minimise the power at point B, hence removing the distortion. This method assumes the PA is the dominant non-linear component. It only works well at a single frequency assuming the modulated band is small compared to the carrier frequency. Wide bandwidth modulated pulses useful in radar systems may upset this. (b) PA + B RF out RF in Canceller 1 Canceller 2 + A Power detector Power detector Figure : Adaptive feed forward system. Unfortunately, because most of these linearization and pre-distortion techniques are derived from ideas used in communication systems, where they are designed for use with communication waveforms. Communication waveforms, although coded, are essentially continuous wave signals. The feedback systems employed provide a continuously varying signal used to control the pre-distortion characteristics to make a power amplifier more linear. Pulsed signals used in radar could not provide the continuous varying signal needed,

12 Ofcom Spectral Efficiency Scheme (SES ) 99 therefore it would be very difficult to make a power amplifier used in a radar transmitter more linear with the same pre-distortion systems used with communication power amplifiers. Communication systems usually have quite a narrow bandwidth when compared with radar. Therefore many of these feedback systems proposed as possible schemes for improving power amplifier linearization will only work over a narrow band. When used over a wider band required by radar systems there is a strong possibility the feedback systems will become unstable. This is because many solid-state devices are needed to achieve sufficient gain. Each device would add a delay to the system that would eventually make the loop unstable when the delay becomes too large. It might be possible to design a system similar to Figure b) based around the linearization technique where instead of a direct feedback loop the PA output is coupled off and sampled by an A-D converter. There would be some digital circuitry that would sample the previous pulse and predict how it needs to pre-distort the RF input pulse for the next pulse. This system could only ever look at the previous pulse and predict how to pre-distort the next pulse. It could not actively change the current pulse. There would simply be too much delay in the sampling and digital process. The current pulse could become corrupted half way through. A digital sampling system may also be designed for use with TWT based transmitters. However the sampling rate would need to be very fast to accurately capture the rising and falling edges of a radar pulse. Even the most modern sampling system techniques might still be too slow to accurately capture the pulse edges without introducing distortions Digital pre-distortion This technique exploits the considerable processing power now available in modern DSP devices. Using modern Direct Digital Synthesis (DDS) techniques, pre-distortion characteristics can now be calculated and a look-up table kept updated. It is most common to pre-distort the signal at base-band or IF frequencies before modulation up to RF frequencies. The most common types of pre-distorter are mapping pre-distorters or constant gain pre-distorters. A mapping pre-distorter uses two look-up tables each of which is a function of two variables (I and Q). This pre-distorter can perform very well except storage and processing times are larger. A constant gain pre-distorter requires only a single dimensional look-up table indexed by the single envelope resulting in a much simpler implementation that requires less memory and processing time for a given performance level Sequential device switching Sequential device switching is a technique that can be used with solid-state radar transmitters or phased array radar (which incorporate separate solid-state amplifiers for each transmitting element and utilise space-combining techniques). If these transmitting elements are sequentially turned on and off then it is possible to build up a shaped pulse waveform formed in space. However, in the case of phased arrays, care must be taken not to destroy the antenna beam shape during switching. This technique allows for all the transmitting amplifiers to operate in a class C mode to maximise efficiency; however there are some very complex timing and phasing issues to deal with. The radar would require a

13 100 Ofcom Spectral Efficiency Scheme (SES ) lot of computing power to generate the correct radar pulse shape and to interpret the return signals Filtering High-power narrow band fixed filters Filters for high peak powers associated with radar are normally waveguide filters i.e. filters built into a length of waveguide. Typically they will have a pass band response, but will also have a finite cut-off frequency due to the waveguide response where there will be zero transmission of RF power. Unfortunately high power waveguide filters are fixed in frequency. These are no good for multi-frequency agile radars such as ATC systems. Mechanically and electronically tuneable filters Mechanically tuneable filters are simply not suitable for radar. All ATC radars have frequency diversity to increase probability of detection. Mechanical tuneable filters are simply not quick enough to respond in time to the speed that radars change frequencies. Electronically tuneable filters may respond fast enough to allow the radar to remain frequency agile; however they simply cannot cope with the power transmitted by ATC radar. Magnetically tuneable filters Magnetically tuneable filters offer low loss and are suitable for high peak powers associated with radars. A common type of magnetic tuneable filter is a kind of E-plane finline filter. Finline filters are ladder shapes printed on a dielectric substrate that spans the broad width of rectangular waveguide. If the dielectric substrate is replaced with a ferrite substrate then the filter can be made tuneable by the application of an external magnetic field. Filters used on the output of radar transmitters to clean up the spectrum can do a good job with out of band and spurious emissions. Their use in clearing up emissions around the fundamental frequency is more limited due to the wide pass-bands required for radar for rectangular waveforms used and frequency agility Solid state Laterally Diffused Metal Oxide Semiconductor (LDMOS) LDMOS devices have been especially useful at UHF and lower microwave frequencies, however devices are now available that operate comfortably as high as S-band. The direct grounding of its source eliminates the inductance associated with bond wires, hence reducing negative feedback that causes the gain reduction at high frequencies. It also eliminates the BeO insulating layer needed in other RF MOSFETs. LDMOS devices also operate off a 28V rail with power outputs of 100W in S-band. At higher frequencies other types of RF transistor are used, e.g. GaAs MESFET based on GaAs and schottky gate junction. GaAs has higher electron mobility than Si, hence is capable at operating at higher frequencies. HFETs and HEMTs improve on the MESFET geometry by separating the schottky and channel functions. The pseudomorphic HEMT

14 Ofcom Spectral Efficiency Scheme (SES ) 101 (phemt) further improves performance by using an InGaAs channel that further increases the electron mobility. LDMOS Technology LDMOS transistors have been replacing bipolar devices in communication base-station applications for a number of years now. Their combination of high power and good linearity make them ideal for this application. These devices could be used for new radar transmitter designs that may require linear amplifiers for production of more spectrally efficient radar waveforms. Advantages of LDMOS Excellent efficiency. This reduces power consumption and cooling costs. LDMOS devices now use a 5th generation 0.4um technology that greatly increases efficiency. These devices typically achieve greater than 30% efficiency. High gain. The LDMOS device utilises a grounded backside source. This substantially decreases the source inductance because there are no bond-wires between the source and package flange. LDMOS devices also have very low feedback capacitance and series gate resistance. Therefore the LDMOS device can have higher gain than existing solid-state RF power devices, typically up to 18dB whilst remaining stable. This high gain will not only reduce the number of amplification stages within the radar transmitter, but also the requirements of the driver devices. Low thermal resistance. Using a grounded source allows the die to be soldered directly to the package flange optimising heat transfer and eliminating the need for toxic beryllium oxide thermally conductive electrically insulation layer used in many RF packaged power transistors. This greatly reduces the cost of the package. The low thermal resistance allows more heat to be dissipated into the heat sink, lowering junction temperatures and therefore increasing reliability. This allows the use of smaller heat sinks and reduces cooling requirement reducing transmitter size and cost. Excellent linearity. Minimising signal pre-correction requirements. High Power Density. Requires fewer transistor packages High Reliability. The Mean Time To Failure (MTTF) figures are excellent. Excellent ruggedness. LDMOS devices do not suffer from thermal runaway due to their negative drain-current temperature coefficient. Benefits of the new technology Due to their higher gain a LDMOS radar transmitter will require less gain stages than bipolar technologies. This will bring a cost saving and increase transmitter reliability. LDMOS devices are increasingly being used in communication systems. As the communication industry moves towards higher frequency bands, devices suitable for radar bands will become cheaper and more suitable for radar use. Major semiconductor maunufacturers already have LDMOS devices on the market aimed at L-band radar designs. Samples for S-band designs are currently available.

15 102 Ofcom Spectral Efficiency Scheme (SES ) Radar transmitters based on different technologies achieve different efficiencies. The best are transmitters based on magnetrons that are generally 50% efficient. Travelling Wave Tubes are typically around 23% efficient. Radar transmitters based on silicon bipolar devices are around 17% efficient after all the combining losses are included. It would be expected that a radar transmitter based on LDMOS technology would achieve a similar figure, but the advantage would be that more spectrally efficient shaped pulses could be used because the LDMOS devices are much more linear. Figure and Figure show transfer curves for some LDMOS devices as test samples or current products. The linearity of LDMOS can be seen when compared with the silicon bipolar transistor shown in Figure The LDMOS devices provide gain for very low-level input powers when biased correctly. Silicon bipolar devices provide no output until the input power is sufficient to switch the device on and therefore are very non-linear until the device is turned on by the input signal. LDMOS gate bias voltage controls its output power. This makes the device suitable for automatic gain control applications and could be used to shape the radar waveforms into a more spectrally efficient pulse. The larger linearity of LDMOS would mean that spectrally efficient pulses would not be corrupted as they pass through the radar transmitter. Control of the gate bias voltage could also be used to turn off the LDMOS devices during the interpulse period, i.e. when the transmitter does not need to amplify a pulse. This would reduce the average drain current and increase transmitter efficiency. A LDMOS radar transmitter would have a similar architecture to previous solid-state radar transmitters presented. The main difference is the power rails needed; LDMOS devices typically require a 28V rail for the drain and another rail for the source. Bipolar devices typically run from a single higher voltage power rail. Replacing solid-state devices in existing radar transmitters with LDMOS would require major redesign work to the power supplies. However the bias point of the LDMOS devices could be set to give the same power output as bipolar devices to minimise changes to the RF architecture. Radar design proving would have to be repeated which is a very costly activity. A new LDMOS radar transmitter would probably be easier to design, but would also be costly. It would have the benefit of exactly the correct architecture for the LDMOS devices and the right power rail requirements. LDMOS technology is maturing all the time. Practical radar transmitters at S-band based on LDMOS technology could be available within the next 5-10 years. LDMOS based radar transmitters will not be significantly more expensive than bipolar based designs.

16 Ofcom Spectral Efficiency Scheme (SES ) Pout (W) Pout 2.7GHz Pout 2.8GHz Pout 2.9GHz Pout 3.0GHz Pout 3.1GHz Pin (W) Figure : Measured transfer curve for a test sample Philips BLS LDMOS transistor. 100W device run at 32V, 200us pulse and 10% duty Output Power (Watts) Input Power (Watts) FH FC FL 1300MHz 1215MHz 1400MHz Figure : Measured transfer curve for a Philips 300YK Silicon bipolar transistor. 300W device run at 48V, 250us pulse and 10% duty.

17 104 Ofcom Spectral Efficiency Scheme (SES ) Figure : Transfer curve from datasheet for a Philips BLL LDMOS transistor. 250W device run at 36V, 100us pulse and 10% duty. Note: curves, (1) f=1.2ghz, (2) f=1.3ghz, (3)f=1.4GHz. In summary solid state LDMOS technology is an excellent candidate for next generation radar transmitter design with advantages in terms of transmitter complexity and reliability as well as significantly improved linearity over the current silicon bipolar technologies, thereby providing the potential for pulse shaping for the purposes of improved spectral efficiency Techniques for new or upgraded radar systems utilising linear transmitters Recent technology advances, leading to the availability of high power solid state LDMOS technology in L-band and its emergence in S-band, means that near linear transmitters become a possibility for radar. With this in mind and an understanding that waveform spectral efficiency is largely controlled by pulse rise (and fall) time an analysis of different pulse rise and fall times was carried out to illustrate the resulting theoretical improvements in spectrum shape and required spatial and spectral separations. Figure illustrates the effect on the theoretical spectrum envelope of a 100µs pulse with rise / fall times varying from the nominal silicon bi-polar time of 168ns up to 20µs. It is clear that spectral width and roll-off rate are functions of rise time with the narrowest spectra associated with the longest rise / fall time. These are tabulated in Table along with the energy contained in the shaped pulses as a percentage of that under a perfectly square pulse envelope this will have an effect on maximum target detection range. Improvements in the resulting radar range and frequency separations are plotted in Figure

18 Ofcom Spectral Efficiency Scheme (SES ) 105 Relative amplitude, db ns 200ns 300ns 500ns 1000ns 2000ns 5us 10us 20us Frequency offset, MHz Rise time Figure : Theoretical spectra for 100µs (trapezoidal) pulse with various rise / fall times Table : Spectral widths for 100µs (trapezoidal) pulse with various rise / fall times Pulse Length (µs) Spectral width (MHz) at : -20dB -40dB -60dB -80dB -100dB -60dB/- 20dB 168ns ns ns ns µs µs µs µs µs Energy (%)

19 106 Ofcom Spectral Efficiency Scheme (SES ) Required frequency separation vs. Range 168ns 200ns 300ns 500ns 1000ns 2000ns 5us 10us 20us Separation, MHz Range, km Figure : Required separations for 100µs (trapezoidal) pulse with various rise / fall times As an alternative, cosine shaped leading and trailing edges or full Gaussian pulses could be produced by a linear transmitter. For comparison Figure illustrates theoretical spectra for a variety of pulse rise / fall times. For these Gaussian pulses, the 100µs pulse width specifies the 3 or 4 standard deviation points of the Gaussian, beyond 100µs the pulse shape is forced to zero. The characteristics of these are tabulated in Table along with the energy contained in the shaped pulses as a percentage of that under a perfectly square pulse. Improvements in the resulting radar range and frequency separations are plotted in Figure

20 Ofcom Spectral Efficiency Scheme (SES ) us trapezoid 20us trapezoid 1us cosine 20us cosine Full Gaussian 3sigma Full Gaussian 4sigma Relative amplitude, db Frequency offset, MHz Figure : Theoretical spectra for 100µs (cosine edge and gaussian) pulse with various rise / fall times Rise time Table : Spectrum parameters for 100µs cosine edge and Gaussian pulses Pulse Length (µs) Spectral width (MHz) at : -20dB -40dB -60dB -80dB -100dB -60dB/-20dB Energy (%) 1µs µs Gauss Gauss

21 108 Ofcom Spectral Efficiency Scheme (SES ) 6 5 Required frequency separation vs. Range 1us trapezoid 20us trapezoid 1us cosine 20us cosine Full Gaussian 3sigma Full Gaussian 4sigma 4 Separation, MHz Range, km Figure : Required separations for 100µs (cosine edge and gaussian) pulse with various rise / fall times For both trapezoidal and cosine edge pulses the 60dB/20dB ratio dramatically reduces (hence the spectrum spread dramatically reduces) with increasing rise and fall times on the pulse edges. This is reflected in the much reduced spatial and spectral separations plotted in Figure and Figure This performance does rely on the amplifier holding the phase of the pulse true to approximately 60dB below the peak. Therefore the power amplifier must not add phase spin as the pulse rises and falls. For target detection, the energy under the pulse is the important factor and any reduction as a result of pulse shaping leads to reduced maximum detection range or Probability of Detection (P D ). For new designs of radar this can be compensated with slightly longer pulses, however, Gaussian shaped pulses with energies as low as 30 40% are not feasible for radar since the pulse lengths would need to be prohibitively long and would likely violate the transmitter duty cycle Improvement compared to traditional silicon bipolar class-c solid state By examination of Table and Table a rise & fall time of 1µs provides a reasonable compromise between energy under the pulse (98.77%) and spectral efficiency thus is a good candidate for a radar system utilising a linear transmitter. In order to illustrate the potential improvements compared to current silicon bipolar class-c solid state technology, the required separations are plotted in Figure for the case of 100µs pulse length with rising & falling edges consistent with current solid state technology with

22 Ofcom Spectral Efficiency Scheme (SES ) ns rise & fall times and candidate linear transmitted trapezoidal and cosine edge pulses with rise & fall times of 1µs Required frequency separation vs. Range 100us NLFM 168ns rise time 100us NLFM 1us rise time 100us NLFM 1us rise time cosine edge Separation, MHz Range, km Figure : Comparison of required separations for class-c vs. linear transmitted pulses A comparison of the required spectral separations at 45km are tabulated in Table Table : Required spectral separations for 100µs class-c and linear transmitted pulses Rise time Shape Separation 45km Difference 45km w.r.t. 168ns rise&fall time pulse Mean difference (MHz) over all ranges 168ns trapezoid µs trapezoid µs cosine A reduction in spectral separation is indicated from 4.2MHz, required for a class-c solid state system, down to 2.5MHz for a linear system with trapezoidal pulses or 1.5MHz with cosine edge pulses for a range separation of 45km, an improvement of 1.7MHz and 2.7MHz respectively. On average, over all ranges, the improvements are 1.5MHz and 1.9MHz compared to the current class-c solid state systems. Pulse edge shaping applied to the short 1µs pulses required for short range detection is less effective since it would reduce the energy under the pulse and thus affect detection performance. However, even modest increases in rise time result in significant spectral efficiency improvements as indicated in Figure and Table

23 110 Ofcom Spectral Efficiency Scheme (SES ) 10 9 Required frequency separation vs. Range 1us pulse 168ns rise time 1us pulse 250ns rise time 1us pulse 250ns cosine edges 8 7 Separation, MHz Range, km Rise time Figure : Required spectral separations for 1µs class-c and linear transmitted pulses Table : Required spectral separations for 1µs class-c and linear transmitted pulses Shape Separation 45km Difference 45km w.r.t. 168ns rise&fall time pulse Mean difference (MHz) over all ranges 168ns trapezoid ns trapezoid ns cosine The 3.8MHz required spectral separation for the 250ns cosine edge linear transmitted pulse, an improvement of 2.9MHz below the 6.7MHz required for a class-c transmitter, is likely to be the limiting case in terms of spectral separation for neighbouring radar at 45km range. Longer rise times would have an increasing impact on detection performance. On average, over all ranges, the improvement is 0.8MHz compared to the class-c transmitter. This is still significantly improved over the TWT required spectral separation of 18.5MHz. It is clear that significant improvements in spectral efficiency can be achieved through the use of linear solid state transmitter technology and either trapezoidal or cosine shaped pulse edges. If spectral efficiency becomes a requirement of future radar systems, then linear transmitter technology using such pulse shapes is likely to be the best approach.

24 Ofcom Spectral Efficiency Scheme (SES ) Summary This sub-section has examined techniques, which could be implemented on existing radar transmitters and also techniques, which could be designed into new radar systems in order to improve radar spectral efficiency. One solution towards better spectral efficiency for the widely deployed TWT based ATC TMA radar (e.g. the Watchman radar) is to change the pulse length. This was explored in section and revealed that some improvements were possible with this approach; however, they were limited by the very rapid rise and fall times of TWT transmitters. In addition, degrading the range resolution in this way may not be consistent with military requirements and thereby not be applicable to a large number of the deployed systems. The move towards solid state transmitters utilising long modulated pulses in ATC radar is consistent with much improved spectral efficiency as a result of the slower inherent rise times and reduced peak powers of these devices. NATS is currently in the process of replacing its deployed radar with modern solid state radar systems, which have these spectral efficiency benefits, however, there are a number of radar deployed in regional airports and MoD sites that still use TWT transmitters. Recent technology advances, leading to the availability of high power solid state LDMOS technology in L-band and its emergence in S-band, means that near linear transmitters become a possibility for radar. Significant improvements in spectral efficiency can then be achieved through either trapezoidal or cosine shaped pulse edges. If spectral efficiency becomes a requirement of future radar systems, then such a solution is likely to be the best approach Conclusions and recommendations This work package has focused on the practical problems of generating high power spectrally efficient waveforms for pulsed radar application. The key component of the radar system for this analysis is thus the transmitter. A number of different high power RF transmitter devices have been investigated and their suitability for transmitting spectrally efficient waveforms assessed. In the key area of medium to long range radar surveillance (which covers both en-route and TMA ATC applications) all historical and current radar transmitter technologies (e.g. Magnetron, TWT and Solid state silicon bi-polar) are class C in operation in order to maintain phase stability and maximise power efficiency. Class C operation, however, is highly non-linear and does not support any kind of pulse shaping (for the purposes of spectral efficiency) beyond the natural rise and fall time characteristic of the devices. A review of candidate spectrally efficient waveforms was also carried out. In the past, many studies of waveforms for radar have neglected the fact that radar transmitters are typically operated in class C and thus are not capable of the linear performance required by some of the proposed waveforms. This review was able to identify those waveforms most suitable for use with class C radar transmitters. It concluded that, while digitally coded waveforms have been proposed for radar application, they would require linear transmitters to achieve improved spectral efficiency. In addition, in practice they can suffer from significant Doppler intolerance and nominally Orthogonal waveforms do not offer sufficiently good orthogonality and time sidelobe performance for radar application. The review of analogue waveforms revealed that NLFM waveforms are amenable to class C transmission and do

25 112 Ofcom Spectral Efficiency Scheme (SES ) offer improvements over the more traditional LFM waveforms and are probably the best option in terms of waveform spectral efficiency for radar application. Spectral efficiency improvement has been assessed based on techniques which allow radars to operate in a way that minimises the interference to radars of a like type, thus improving spectral efficiency by reducing the spatial or spectral distance between neighbouring radar. This allows a network of radars to operate over a narrower range of frequency allocations. As a result of these considerations a number of conclusions and recommendation for current technologies can be made: Given the analysis in section it is clear that co-axial magnetrons offer significantly improved spectral efficiency compared to traditional magnetrons. For magnetron radar, changing to co-axial magnetrons should thus be considered when maintenance / upgrade intervals allow. One solution towards better spectral efficiency for the widely deployed TWT based ATC TMA radar (e.g. the Watchman radar) is to change the pulse length. There is scope within the NATS criteria for the Watchman radar to operate with a longer pulse length and still meet range resolution and range accuracy requirements. This was explored in section and revealed that some improvements were possible with this approach; however, they were limited by the very rapid rise and fall times of TWT transmitters. The spectral efficiency of modern solid state radar transmitters utilising long NLFM pulse waveforms is close to the best spectral efficiency of current radar technologies. However, the silicon bipolar solid state technology utilised in these transmitters is limited in its maximum achievable rise time to less than 200ns. Since rise (and fall) time to a large extent control the shape of the transmitted spectrum, this limits the achievable spectral efficiency of this type of transmitter. In terms of the next generation of radar transmitters: Recent technology advances, leading to the availability of high power solid state LDMOS technology in L-band and its emergence in S-band, means that near linear transmitters become a possibility for radar. Although LDMOS technology is different to Silicon bi-polar and in order to fully realise its benefits, new transmitter design must be undertaken, it is well understood and development timescales are such that practical radar transmitters could be developed within 5 years. This is the most likely and low risk direction for the development of the next generation of radar transmitters. With the possibility of near linear radar transmitters and an understanding that waveform spectral efficiency is largely controlled by pulse rise (and fall) time an analysis of different pulse rise and fall times was carried out to illustrate the resulting theoretical improvements in spectrum shape and required spatial and spectral separations. This indicated that significant improvements in spectral efficiency can be achieved through either trapezoidal or cosine shaped pulse edges, with the longer rise times giving the best improvements. If spectral efficiency becomes a requirement of future radar systems, then such pulse shapes are likely to be the best approach.

26 Ofcom Spectral Efficiency Scheme (SES ) 113 It should be remembered that most analysis in this and other previous reports has mostly been carried out based on theoretical spectra. As illustrated in section , real transmitted spectra, even of solid state transmitters, are modified by various real world factors which are not commonly modelled at this level of detail. Thus the transmitted radar spectra achieved through deployment of new radar technology must be modelled in more detail and/or measured before any allocation re-planning can take place.

27 114 Ofcom Spectral Efficiency Scheme (SES ) This page is intentionally left blank.

28 Ofcom Spectral Efficiency Scheme (SES ) A review of the degree to which frequency and time agile applications can co-exist with radar in the same band without causing interference and the effect of secondary users Introduction The original driver for this work was the Cave Independent Audit [39] which suggested that studies should be carried out into the feasibility of gaining more spectrum by sharing the spectrum with radars in the S band. In this section the possibility of sharing bandwidth in both the L and S bands is examined. The primary users of these bands are air traffic control (ATC) radars and military radars. The potential secondary users of these bands were identified generally as civilian wireless communication systems, including cellular mobile radio, wireless LANs and fixed wireless access systems. In order to meet the demand for more spectrum for these systems three questions are considered: i) What are the performance and quality of service (QoS) tradeoffs due to multiple occupancy of the spectrum, mutual interference and the growth of the noise floor? ii) iii) What are the methods of achieving mutual occupancy of the spectrum, including smart modulation and coding, adaptive interference cancellation and exploitation of the spatial domain? Is a multi-static radar approach more attractive than a mono-static approach for maximising multiple occupancy through the exploitation of space diversity? In the context of this report, bandsharing is assumed to refer to the situation where a secondary user occupies bandwidth that is licensed to a primary user within the same general geographic area as the primary user on a mutual non-interference basis. The services provided by the primary and secondary users need not be the same; indeed, they will most often be totally different, exploiting the orthogonality that exists between the two different systems needed to support those services. In this report, the term bandsharing does not refer to the situation where the primary user and secondary user may be in totally different geographic areas (where potential interference is minimal), even though the license for the primary user may include these regions. Furthermore, this study does not consider how existing wireless standards such as GSM and 3G will coexist with the L and S band radars, since their designs and signal sets have been optimised to provide best performance within licensed spectrum for which the interference characteristics of their respective environments is fixed and well known. They possess neither the smartness nor the agility to operate effectively in the proposed bandsharing environment. However, because of the localised nature of wireless LANs defined by the x (WiFi) standard [40] and emerging standards such as WiMAX e and h [40], these are possible candidates for operation in the L an S bands. Technological developments in active spectrum management, software defined radio (SDR); and cognitive radio in particular, have brought about a paradigm shift in the way that spectrum management can be handled and this has been recognised by Ofcom and the

29 116 Ofcom Spectral Efficiency Scheme (SES ) FCC. Indeed next generation radio systems such as 4G, in the case of cellular mobile radio, Wireless Regional Area Networks covered by standard and ultrawideband communication technology defined under the Recommended Practice [40] and the Framework Particulars [40], may well incorporate active spectrum management techniques that allow coexistence with primary users in shared frequency bands. Consequently, the methods that may be adopted by such emerging standards do play an important part within the context of this report. The background to this change in approach to spectrum management can be found in the more liberal attitudes being taken by both Ofcom and the FCC over the past decade to usage of the licensed frequency bands, which have been largely technology driven. This recognises that managed spectrum, particularly in the USA, is used extremely inefficiently with less than typically 16% [41] - [44] of the managed spectrum being occupied, even at busy times. Software defined radio, in conjunction with adaptive modulation techniques that might include both spread-spectrum and OFDM, and sensor technology that can sense the RF environment in the vicinity of the radio system (so-called cognitive radio [45] - [48]) is beginning to emerge; it allows fully adaptive wireless technology that is capable of adapting to the RF environment in terms of operating frequency, power level and waveform with the objectives of maximising spectrum usage without degrading the quality of Service (QoS) to any of the users of the band. In addition to the work of the Cave audit for OFCOM, the FCC has issued a number of Notices of Inquiry (NOI) and Notice of Provisional Rule Making (NPRM) that will facilitate multiple occupancy of the frequency bands with a view to gaining more spectrum within the USA, including: ET Docket No on Authorization and use of software defined radio ET Docket No on Additional spectrum for license-exempt devices below 900MHz and in the 3GHz band ET Docket No (NOI) on Receiver Standards ET Docket No Spectrum Policy Task Force Report ET Docket No (NPRM/NOI) on An interference temperature metric to quantify and manage interference and to expand available unlicensed operation in the fixed, mobile and satellite frequency bands ET Docket No (NPRM) on Facilitating opportunities for flexible, efficient and reliable spectrum use employing cognitive radio technologies ET Docket No on License-exempt operation in the TV broadcast bands In particular, ET Docket Spectrum Policy Task Force Report [42], concluded that new smart radio technologies would enable better and more intensive access to the spectrum (including the concepts of bandsharing, multiple spectrum occupancy and dynamic spectrum management). As a result of this more liberal approach to spectrum management, research in the general area of bandsharing has been gaining pace over the past three years. Since many of the proposed techniques to facilitate more intensive use of the spectrum are relevant to the work on bandsharing proposed by Ofcom for this project, it

30 Ofcom Spectral Efficiency Scheme (SES ) 117 is timely and relevant to review the most appropriate of the new techniques here and to relate their relevance to this project s aims. The details of this review of smart radio technology to facilitate bandsharing are provided in Annex A Methods of spectrum sharing Generally, where bandsharing is proposed between a licensed primary user and an unlicensed secondary user, it is usual to assume that the methods adopted to provide mutual protection of both types of user are provided by the secondary user, since it is presumed that the standards are currently in place for the existing primary user. In these circumstances, the aim of all bandsharing schemes is for the secondary user to adopt a waveform that is as close as possible to being orthogonal to the waveform of the primary user. In order to achieve this, the secondary user may exploit the orthogonality that exists in any or combinations of: i) the frequency domain (by choosing to operate in parts of the allocated frequency band that are temporarily unoccupied by the primary user), ii) iii) the time domain (exploiting the fact that the primary user is not currently transmitting and the primary user receivers are not expecting to receive at that instant), or the correlation domain (by using modulation schemes and waveforms for primary and secondary users that have minimal cross-correlation), so that the systems may transmit simultaneously and on the same frequency. Adaptive signal processing may be used to provide additional interference rejection for each of these techniques and hence maximise mutual protection. In addition, exploitation of the spatial domain can provide enhanced mutual protection. This may include operating in different geographic areas where mutual interference is low, exploiting naturally occurring or artificially generated shadow regions, or using directional antennas and adaptive interference nulling techniques to minimise mutual interference. Loss of orthogonality between the waveforms of the two users results in an increase in interference both to the primary user and the secondary user, which directly affects the range at which the primary (and secondary) signals are detectable, and hence the coverage areas. The methods used to provide minimum levels of mutual interference, described above, are only one way of classifying a bandsharing system. An alternative classification is the way in which the bandsharing policy is applied. There are three broad approaches to bandsharing policy: underlay, overlay and interweaving Underlay method In the underlay method, the secondary user generally adopts a modulation scheme that has noise-like characteristics as far as the primary users are concerned. A typical example might be the case of an ultrawideband (UWB) direct-sequence spread-spectrum signal or a multi-carrier random-phase signal operating with a very low power spectral density, whose spectrum is superimposed on the spectra of the primary user (whose spectrum is

31 118 Ofcom Spectral Efficiency Scheme (SES ) narrowband in comparison) Figure shows the case of an ultrawideband system underlaying an WLAN system. Power density (W/Hz) Not to scale GSM WLAN (primary user) GPS WLAN bluetooth Increase in noise floor for primary user Very low psd of secondary user UWB radio (secondary user) Frequency (GHz) Figure : Schematic of the underlay approach to bandsharing In this approach, the excellent auto-correlation properties of the ultrawideband signal are exploited to despread the ultrawideband signal to recover the energy in the data [49] - [51]. At the same time the low correlation assumed to exist between two signal types ensures that the interference at the secondary receiver due to the WiFi system is acceptably small. Furthermore, the low PSD of the noise-like signal of the UWB system results in a small increase in the apparent noise floor for the primary user receivers in the vicinity of the secondary user s transmitter. The effectiveness of this method depends on (i) the degree to which the two types of waveform can be made orthogonal and (ii) the service area required by the secondary user (which sets its transmitter power). This in turn sets the growth in the interference floor for the primary user, which ultimately reduces the detectable range of primary user signals. The impact of this depends critically on the relative location of primary and secondary users. Not surprisingly, handset geolocation is beginning to play an extremely important part in the mitigation of interference from cognitive radios [52]. The advantage of this method is that it does not rely on using smart communication technology for it to be effective, but it is essential to use the most appropriate waveforms to minimise signal correlation between the primary and secondary users and it does require the secondary user to make use of ultrawideband technology. In addition, the coverage areas for the primary and secondary users need to be managed very carefully in order to make sure that the transmitter power of one of the users enables the other user to have an acceptable coverage area. Not surprisingly, this type of bandsharing policy was one of the first to be considered and early work by Ormondroyd and Shipton [53] - [57]from 1978 onwards considered how very wideband (25MHz) direct-sequence spread-spectrum underlays could be used to provide additional communication services by sharing with the TV bands in the UK on a mutual non-interference basis. Their work included detailed studies of the effect of spread-spectrum interference on the subjective impairment of TV signals and the impact of the TV transmissions on the reception of the spread-spectrum signals. From this they were able to establish the required protection margins that both

32 Ofcom Spectral Efficiency Scheme (SES ) 119 primary and secondary needed to operate and using this they were able to establish service areas where bandsharing was possible Theoretical model of the underlay technique Co-sited transmitters forward link To illustrate the underlay technique, consider the case of the forward link of two basestation transmitters that are co-sited. One of them is the primary user and the other represents the secondary user. Let the primary user be a conventional narrowband system of bandwidth W p and the secondary user be a wideband spread-spectrum user of bandwidth W s >>W p. Assume that the data rate for both systems are the same, W d. Also assume that the spreadspectrum waveform is noise like as far as the primary user is concerned. The path loss is L p, and assumed to be the same for both the primary and secondary systems. Let the spreading factor for the direct sequence spread-spectrum system be G = W W. The transmitter power for the primary user is P p and for the secondary user it is P s. Let the antenna gains of the primary and secondary transmitters be and and the receiver antenna gains be and respectively. G rp G rs In the absence of any interference from the secondary user, the noise power at the receiver of the primary user is: N p = ktfw d, where F is the system noise figure referred back to the detector. If the required SNR for adequate detection of the data is SNR0 for both primary and secondary users, then the minimum received power must be: G t p p G ts s d P rec = SNR0 N p min Equation The service area is defined by the maximum acceptable path loss to provide the necessary P rec min which is obtained from the link budget equation L pmax Prec SNR N min 0 p = Pt GtGr Pt GtGr = Equation For example, for propagation and plane-earth multipath loss, given by ht and by: hr Lp = ( h h ) 2 t r 4 r, where are the transmitter and receiver heights, the radius of the service area is given r max p = 4 P p G t p SNR G 0 r p ( h h ) t ktfw r d 2 Equation However, with the addition of the secondary user, the primary user receiver now sees increased interference due to the spread-spectrum interference that falls within the detector s bandwidth. This is:

33 120 Ofcom Spectral Efficiency Scheme (SES ) ( ) 4 2 r G h h G P G ktfw N p r t t r s d p s p + = Equation and is reduced to: r max ( ) ( ) max d p r t t r s r t r t p FkTW SNR G h h G G P SNR h h G G P r s p p p p = Equation subject to the constraint that: ( ) ( ) > p r t t r s r t r t p G h h G G P SNR h h G G P s p p p Equation For the secondary user, the total interference is given by: ( ) + = 4 2 r G h h G G P ktw N p t r t r p d s p s Equation and the corresponding service area for the secondary user is ( ) ( ) max d p r t t r p r t r t s kftw SNR G h h G G P SNR h h G G P r p s s s s = Equation subject to the constraint that ( ) ( ) > p r t t r p r t r t s G h h G G P SNR h h G G P p s s s Equation Thus from these equations we see that, if the antenna parameters are the same for both primary and secondary users and we allow the secondary user and primary user to transmit at the same power, the service areas for both users are the same and less than if secondary user were not transmitting. The reduction in the service area when the secondary user is present is dependent upon the spread-spectrum process gain (i.e. spreading factor). The reason for this is that when the process gain (spreading factor) is high, for a fixed transmitter power, the power spectral density of the spread-spectrum signal is low in inverse proportion to the spreading factor, and this reduces the level of interference falling within the primary user s bandwidth. By the same argument, the interference from the primary user that the secondary receiver sees within its detector bandwidth is also reduced for high spreading factors; since the G p

34 Ofcom Spectral Efficiency Scheme (SES ) 121 despreading process in the spread-spectrum receiver spreads the power of the primary transmitter over the spread-spectrum transmit bandwidth, where its power is reduced by the narrow pre-detector filter or correlator. For the case where the primary transmitter power is larger than the secondary transmitter power, we see that the service area for the primary user is increased, whereas the service area for the secondary user is correspondingly reduced due to the increased interference from the primary user. When the primary transmitter power is sufficiently large that the constraint of Equation is not met, the secondary user has no service area because it has insufficient process gain to counteract the large interference level from the primary transmitter and underlay bandsharing is no longer possible. This has an impact when attempts are made to bandshare low power secondary user systems with high power primary user systems. As an example, assume that the primary user is a P t kw omni-directional broadcast transmitter and the secondary user is a 1W spread-spectrum transmitter co-sited with the broadcast transmitter with a x1000 spreading factor. Neglecting near field effects, and assuming all antenna gains to be equal, the service area of the secondary user collapses to zero when the power of the broadcast transmitter exceeds Pt = G p SNR0. If we assume that the minimum value for SNR0 is typically x10, then the maximum power allowed for Pt is less than 100W if the secondary user is to be able coexist, and even then the service area is vanishingly small. For this type of bandsharing system to work it is vital to use every possible technique to minimise the interference. Using spread-spectrum process gain (spreading factor) is not ideal because, for the bandsharing case, the process gain not only extracts the signal from the other user interference; it also defines the allowable dynamic range of the two signals, as in the case above. Consequently it is usual to use spatial filtering in conjunction with spectrum spreading to provide the necessary isolation between the primary and secondary users. If the primary user is a high power micro-wave point-to point link and the secondary user is a low power local loop fixed wireless access system, such that the gain of the primary transmit antenna in the direction of a secondary receiver is = -10dB and the secondary user s receiver gain in the direction of the primary user transmitter is -10dB, then the maximum allowable transmit power for the primary transmitter increases to 10kW before the secondary system ceases to be able to provide a service. These equations have some implications for the case where a high power radar is being used in conjunction with an underlay bandsharing system, since the feasibility of cochannel operation rests almost entirely on the ability to reduce interference through spatial filtering. G t p Co-sited transmitters reverse link The situation for the reverse link, where many primary and secondary transmitters may be operating within the allowed service areas, is much more complex to analyse than the previous case. Not only must each secondary spread-spectrum handset be power controlled to ensure that all handsets can be adequately received by the spread-spectrum basestation, but the presence of secondary user handsets close to the primary basestation receiver can cause unacceptable levels of co-channel interference to the spread-spectrum base-station. This is exceptionally difficult to combat.

35 122 Ofcom Spectral Efficiency Scheme (SES ) For the case of a distribution of M secondary handsets causing interference to the primary user s basestation receiver, the reverse link coverage area is defined by: ( h h ) 2 PT GT G P p rp t r r = 4 M PT GT G si si RPi t SNRo ktfwd + i= 1 G p 2 ( h h r ) r i 2 Equation where, r i is the range of the i th spread-spectrum handset from the basestation, transmit power of the ith spread-spectrum handset, is the gain of the i th spreadspectrum handset antenna in the direction of the basestation and G Tsi P Tsi primary user antenna in the direction of the i th spread-spectrum handset. G Psi is the is the gain of the This can be evaluated for specific instances of handset distributions, or it is possible to assume average distributions of handsets and solve by integration. Generally, the coverage area for the reverse direction is worse than for the forward direction unless all secondary users are located at the very edge of the primary user service area because of the presence of high power interference for handsets close to the basestation. A similar type of expression can be evaluated for the coverage area of the secondary user s reverse link. Again the service area is generally worse for the reverse link under bandsharing conditions because the primary user handsets are not power controlled in the same way that the secondary user handsets are power controlled, and these cause high levels of interference when they are close to the secondary user s basestation. Non-cosited transmitters forward link A similar situation occurs when the primary and secondary basestation transmitters are not co-sited. In this case, the maximum range of the worst-case primary receiver from its basestation is: rmax p ( h h ) 2 PpGt G p rp t r = 4 Ps Gr G p ts t SNR0 ktfwd + 4 G prsp ( h h ) 2 r Equation where r sp represents the range of the secondary transmitter to the primary receiver of interest. A similar type of expression can also be obtained for the allowable coverage area for the secondary system and also for the reverse links for both cases. The case of non-cosited base-stations produces far worse results for coverage area than the co-sited case considered earlier. To see this, consider what happens to as r sp becomes close to the handset defining the coverage area. Figure illustrates one scenario that arises from this situation. Here the range from secondary basestation to a primary handset is reasonably large and the required coverage of the secondary service is defined here as being small. Consequently, the required secondary user basestation r max p

36 Ofcom Spectral Efficiency Scheme (SES ) 123 transmitter power is quite low and the level of interference at the primary handset is quite low resulting in a relatively small reduction in service area. However, if either the required service area for the secondary service increases, or the secondary user is closer to the primary handset, the coverage area for the handset shrinks considerably. Primary user Rx at edge of service area secondary basestation Primary user Rx at edge of reduced service area Primary BS Defined coverage area for primary users Primary BS New defined coverage area for primary users secondary coverage area Original coverage area Figure : Simplified schematic of a bandsharing scenario using underlay Interference temperature The FCC Spectrum Policy Task Force 3 has recognised the importance of the underlay method of bandsharing. Based on calculations similar to those given above, it has considered the benefits of allowing unlicensed users access to licensed bands using underlay bandsharing by allowing a slight growth in the prevailing noise floor due to the spatial distribution of secondary users. It will be seen from Equation , that the presence of the secondary handsets effectively increases the noise floor as seen by the primary users. The method that the FCC have used is to represent the increase in interference power due to the secondary users is via the increase in the equivalent noise temperature 4 of the environment local to the licensed users. It is defined as the interference temperature [58] - [62]: Ti N N + Ii i=1 = kw Equation where N is the measured noise power within the equivalent noise bandwidth of the wanted (i.e. primary user receiver), is the interference power of the i th secondary user in the I i 3 ET Docket No Spectrum Policy Task Force Report 4 ET Docket No (NPRM/NOI) on An interference temperature metric to quantify and manage interference and to expand available unlicensed operation in the fixed, mobile and satellite frequency bands

37 124 Ofcom Spectral Efficiency Scheme (SES ) vicinity of the primary receiver of interest that falls within the noise bandwidth, W, of the primary user of interest and k is Boltzmann s constant. This concept has led to considerable debate, since it makes assumptions about the statistical nature of the interference that may not be true (i.e. that the interference is Gaussian distributed). If the secondary user is a multiple access spread-spectrum system that uses scrambling codes and the central limit theory applies because of the number of interferers, this may be a valid argument. However, if the number of interferers is relatively small and their characteristics are not noise-like, this may lead to considerable underestimates of the impact of the interference on the primary user. Its advantage is that it is simple to use and is based on real-time measurements of the interference power that falls within the bandwidth of the receiver of interest FCC Concept of Underlay Bandsharing The FCC concept of underlay bandsharing is similar to that discussed above. It proposes that certain licensed services, which currently enjoy a particular service area under the prevailing noise floor, will relinquish the guarantee of some of that service area for the greater good by allowing a limited increase in the noise floor over the entire service area, as shown in Figure by allowing unlicensed bandsharing. The maximum level of interference floor will be capped (although precisely how is debatable). If no one shares the bandwidth, the service area of the primary user is not changed, but as the number of bandsharers increases and the noise floor rises to the capped level, the service area of the primary user shrinks. Power at Receiver licensed signal Prevent aggregation above interference temperature limit New Opportunities for Spectrum Access Service range at original noise floor Guaranteed service range with interference cap Noise Floor Current FCC Power Limits Distance from licensed transmitting antenna Figure : Interference model of underlay bandsharing taken from FCC ET Docket Figure below shows the underlay concept applied to ultrawideband technology. This figure emphasises how there might be different allowable interference temperature limits depending on the priority of the primary user (e.g. GPS, which demands that the interference floor is significantly reduced in frequencies around the GPS carriers because of its low power).

38 Ofcom Spectral Efficiency Scheme (SES ) 125 Figure : Schematic of the underlay approach to bandsharing highlighting the need for different levels of interference temperature cap to protect vulnerable or high priority primary users via an interference mask Drawbacks of the underlay method Whilst the advantages of allowing greater access to the spectrum is self evident and the level of intelligence required by the secondary bandsharing systems is relatively limited, there are a number of drawbacks to underlay methods. First, there is an increase in the background interference level, which affects the existing systems either through reduced service area or reduced capacity. In addition, certain non-communications users of the RF bands such as radio telescopes etc, may be severely affected by even small increases in the interference noise floor. Second, the level of fairness to the primary licensed users rests in the ability of the bandsharing systems to self limit their interference to the prescribed interference noise floor cap for that band of frequencies. This requires that the secondary users are fitted with interference monitoring sensors so that they can automatically reduce their power levels to maintain the guaranteed performance for the licensed users. However, the noise floor is set by many interferers, not just one, so although individual users monitor the local interference environment, maintaining the level of interference to prescribed limits will require collective action. This means that maintaining the noise floor will require some form of federated approach whereby all the secondary users are linked to provide a reliable interference model. Third, dynamic power control of the secondary users may result in serious transient service area problems for the secondary user. Fourth, the fluctuation in interference will be a dynamic process as secondary users enter and leave the service area. This will affect the accuracy of the interference measurements and may lead to a form of dynamic instability if the averaging time needed to estimate the interference temperature is of the same order of magnitude as the coherence time of the

39 126 Ofcom Spectral Efficiency Scheme (SES ) interference. Finally, the interference temperature mask needs to be defined and set in legislation. Since this approach to bandsharing is heavily dependent upon cross-correlation between the primary and secondary user and any imperfect cross-correlation reduces the dynamic range of the signal by locally increasing the interference floor, it is vital that spatial filtering through adaptive beam forming and spatial nulling are used wherever possible to reduce the interference before the correlator needs to process the resulting interference signal. In some situations, for example bandsharing a fixed wireless access network with a fixed point-to-point microwave link, this is very convenient. However if both primary and secondary users hope to establish omni-directional broadcast communications the technique is not applicable Overlay method The overlay method is based on dynamic allocation of spectrum to the secondary users. This approach is based on the observation that the actual spectrum occupancy of licensed spectrum is generally low. By monitoring the bands of interest, the secondary users are able carry out a free band search and occupy those sections of the spectrum that are currently white (i.e. unused). This approach requires a totally different type of waveform to the underlay method. Clearly, given the dynamic nature of spectrum usage, the likelihood of finding large contiguous bands of unused spectrum at any given time is much smaller than finding lots of narrow bands of unused spectrum. Consequently, it is advantageous for the secondary user to operate with relatively narrowband signals in order to maximise its ability to locate and occupy narrow bands of white spectrum. An optimum type of waveform for the overlay approach is a multi-carrier waveform, in which each sub-carrier is narrowband and the frequency of each of the sub-carriers can be selected at will to perfectly match the white spectrum. In this way, the overall capacity achievable by the secondary system can be quite substantial and scalable if the secondary user can identify and occupy sufficient free channels. An obvious candidate waveform has its origins in orthogonal frequency division multiplexing (OFDM) [64] - [69]. A schematic of this concept is shown in Figure Here, the primary user channels are separated by different width frequency guard-bands and these have been filled by the dynamic allocation of different numbers of OFDM sub-carriers in each guard band [70]. In OFDM, blocks of N complex-valued data symbols corresponding to the modulation scheme being used (e.g. M PSK or QAM) are used as the inputs to an inverse FFT algorithm of length N whose output is N complex-valued tones corresponding to the pattern of symbols in the input block 5. The position of each symbol within the block determines the frequency of the tone that that symbol is carried on. Setting a symbol to zero ensures that the power in that tone is also zero. Consequently, loading symbols into appropriate positions within the input block and setting the other positions in the input block to zero determines the spectrum of the OFDM signal 6. 5 By replacing the IFFT with an inverse discrete cosine transform, real tones are generated 6 This is bit loaded OFDM

40 Ofcom Spectral Efficiency Scheme (SES ) 127 primary users Signal power secondary users secondary users frequency Figure : Schematic of the overlay concept of bandsharing Obvious applications of this approach are bandsharing using the frequency guardbands between TV transmissions to prevent the effects of adjacent channel interference. This is clean and unused spectrum. However, it is there for a purpose, and so it is essential that bandsharing systems do not introduce unacceptable levels of adjacent channel interference to the TV receivers requiring well defined spectra. Here again, OFDM systems could provide the solution since their spectrum is generally well defined. However, in order to achieve this, they do need highly linear power amplifiers. The overlay method requires transmitters and receivers that can monitor the frequency bands of interest for activity by the primary user. This is a practical proposition using modern digital signal processing. Having identified free-bands, appropriate bit loading assignments are used to dictate the precise sub-carriers of an OFDM system (for example) that can be used in a particular geographic area to best fill the unoccupied spectrum. A protocol then needs to exist that advises secondary receivers which frequencies the transmitters have decided to use. One possible approach to this is to use one of several possible coordination channels that the receivers constantly monitor. The situation of Figure does not indicate the dynamic nature of the situation and Figure [72] shows a schematic of a hypothetical time-frequency plot, where the occupied spectrum is shown as the coloured regions and the unoccupied spectrum as white, which is being seized by the secondary users using dynamic spectrum access (DSA). Also shown is a black region where the licensing authority has mandated that the frequency band is protected. The Figure illustrates: i) how secondary user 1 attempts to create a single contiguous channel by first occupying the free channel f 2 and then continues transmitting on f 4 when that channel becomes free, ii) how it is able to allocate itself more bandwidth into f 3 and then hops to f 2, when f 3 becomes blocked to it, and iii) how secondary user 2 uses DSA to occupy f 1 but loses the contention battle to hop to f 4 (which is won by secondary user 1) and has to wait to continue the transmission on f 1 when that frequency becomes free again.

41 128 Ofcom Spectral Efficiency Scheme (SES ) Intelligent overlay bandsharing systems will keep a log of white spectrum and will advise the corresponding receiver of these frequencies. Consequently, when one block of frequency is lost, it is immediately able to alter the bit loading pattern to start transmission on a new pattern of sub-carriers so that the transmission of data is contiguous even though the frequencies used are hopping over the band. In this sense, the bit-loaded OFDM system can be viewed as a form of agile multi-carrier frequency hopped system. A number of issues must be addressed. A key issue is the method by which the spectrum is searched for free spectrum. Since the utilization of the spectrum is dynamic, the spectrum needs to be assessed quickly to ensure that valuable spectrum usage is not wasted during the monitoring process. However, there are key tradeoffs between the speed of assessment, the amount of spectrum that must be searched and the required sensitivity of the assessment. This is a particularly difficult issue since the sensor on a particular secondary user transmitter is only able to sense the RF environment locally (dependent upon the sensitivity of its sensors). What that sensor perceives as free spectrum may, in fact, be very low level emissions from a distant transmission which has a well established link with its own local receiver. This is the DSA equivalent of the hidden node problem. In this case, marking spectrum incorrectly as white may cause an unacceptable level of interference to the receiver of the primary user. Consequently, this type of approach requires a fully coordinated approach to spectrum monitoring across the entire spectrum of interest gained from many secondary users in different geographic locations, so that all user share a common time-frequencyspace table. f 5 Excluded spectrum f 4 DSA frequency f 3 f 2 DSA DSA DSA D S A f 1 DSA DSA time Figure : Schematic of dynamic spectrum allocation technique showing intelligent exploitation of time and frequency plane A further important question for dynamic spectrum allocation systems is the optimum granularity in the time domain. There are far more opportunities for achieving relatively short durations of white spectrum than long periods. However, this must be traded against the minimum time needed to observe the spectrum to an acceptable level of sensitivity in

42 Ofcom Spectral Efficiency Scheme (SES ) 129 an attempt to minimise the hidden node problem. Equally importantly in the choice of timeslot granularity is the structure of the data frames used by the secondary systems, which is dependent on the protocols used by the system. If too fine a time slot structure is used, the benefits of greater access opportunities to the spectrum are negated by the poor payload to overhead ratio of the short time frame. Whereas a much coarser time granularity offers a better ratio of payload to overhead but the opportunities for finding free slots of that duration are much fewer. Another important issue is one of bandsharing etiquette. Having occupied the frequency band, the secondary user must quit the band as soon as the primary user requires it otherwise the primary user may face unacceptable co-channel interference. Consequently, the secondary users must constantly monitor the bands and use waveform recognition techniques to establish when the primary user requires the band. Even so, a minimum lag must occur before the secondary user recognises that the primary user has returned to use the spectrum and this may cause unacceptable interference to the important headers that normally precede the data packet. This is a key issue that must be resolved for this approach to be acceptable A cognitive radio approach to bandsharing One possible solution to this problem is the use of cognitive radios [43] - [48], which are described in greater detail in Annex A associated with this work package. A cognitive radio will have a range of waveform sets that it can draw upon depending on the circumstances. One of these waveforms would certainly be an OFDM waveform for use where overlay bandsharing is the best approach to gaining access to the spectrum. However, it may also include a UWB mode or a DS spread-spectrum mode for when underlay is the most appropriate bandsharing strategy. It also has sensors that monitor the RF environment. These sensors will check for: (i) free spectrum, (ii) the local interference environment and (iii) actual data transmissions across the band. It may also be linked to other cognitive radios operating in the area so that a wider picture of the RF environment can be obtained, as discussed above. The cognitive radio will also have the ability to learn from the past usage of the spectrum to predict future usage patterns 7. In this way, a cognitive radio system may be able to predict when a primary user is next likely to require access to the frequency band again based on its previous pattern of spectrum usage so that it can release the spectrum to the primary user very quickly, thereby minimising interference. Central to this approach of dynamic spectrum allocation is the need for the secondary users to exploit the available white spaces fairly and in more intelligent dynamic spectrum algorithms this will play an important part in deciding how spectrum is allocated. As illustrated in the Figure above, contention for free spectrum by the various secondary users will occur and one of the ways of resolving contention could be on the basis of which user had been allocated the most resources prior to the occurrence of the contention. Again, since cognitive radios are likely to share information concerning the RF environment, they 7 It has been suggested that some types of cognitive radio will contain complex models of the RF environment. The sensors will provide the data needed to populate this model, which will form the basis for traffic predictions

43 130 Ofcom Spectral Efficiency Scheme (SES ) can also share information about their own QoS requirements, their priority status and the net throughput of data that each user has currently enjoyed. Consequently, cognitive radios could be used to collectively decide contention issues as well as how spectrum is to be accessed fairly. Questions arise, however, as to precisely how a set of cognitive radios can collectively and cooperatively optimise the use of the white spectrum dynamically. The problem is a complex multi-objective optimisation problem [73], where the primary objective is not to cause unacceptable levels of interference to the primary user and the secondary objective is utilise as much free spectrum as possible for the secondary users. Another objective will be to choose spectrum that attempts to meet each user s QoS requirements. Techniques such as multi-objective optimisers based on evolutionary algorithms [74] - [76], or Tabu search methods [77] - [80] or even gaming strategies [81] - [82], may provide solutions to this problem. However, the question remains as to whether the optimisation process is handled centrally or is decentralised and whether the processing is fully decentralised or federated Active power management and handset geolocation Overlay bandsharing systems can exploit the opportunities that exist due to the normal geographic frequency reuse that many systems use (including the TV bands and GSM cellular radio, for example) to minimise co-channel interference. However, it is important to recognise from the outset that in exploiting these spectrum opportunities, the bandsharer must not be the cause of co-channel interference. Consequently it is important that overlay bandsharing methods incorporate active power management to ensure that the power used is the minimum needed to establish the link to the required QoS without causing (cochannel) interference to the primary users. As referred to earlier, this is a difficult problem because the primary receivers are effectively hidden nodes and it is difficult for a secondary user to assess the level of interference that its transmission is having on receivers at a distance. It is clear, however, that the likelihood of a secondary user inadvertently generating co-channel interference for a primary receiver will be minimised if the required service areas of the secondary wireless networks are small in relation to the coverage area of the primary service area, since the power requirements of the secondary links are minimised. Although active power control is now an established technique for direct sequence spreadspectrum systems such as CDMAone and W-CDMA, it is believed that the tolerance limits required for the power control algorithms for secondary-user bandsharing are much more stringent than for primary-user bandsharing, such as W-CDMA. Consequently, it has been argued that future overlay systems based on cognitive radios will require accurate handset geolocation techniques so that link margins of all the users in the secondary network can be accurately assessed to minimise co-channel interference. However, such suggestions seem to ignore the significant impact that multipath fading would have on the predicted link budgets.

44 Ofcom Spectral Efficiency Scheme (SES ) Interleaved bandsharing heteromorphic waveforms This type of approach shares the benefits of both underlay and overlay approaches and maximises the opportunity for gaining access to the spectrum. In this approach, both white spectrum and grey spectrum is utilised. Grey spectrum is occupied spectrum. Figure below, taken from Seidel [72] at Raytheon illustrates the concept. Here, opportunities for utilising white spectrum are taken using overlay methods based on OFDM and further bandwidth is gained by using ultra-wideband or direct-sequence spread-spectrum underlay methods spread across several discontinuous bands of relatively narrowband users. It is particularly important that the underlay is performed with primary waveforms that have low cross-correlation in order to minimise the degree of cross-correlation between primary and secondary users. The composite waveform set used by the secondary user is referred to as a heteromorphic waveform because (i) different, waveforms occupying discontiguous blocks of spectrum are used by the secondary user to provide a single contiguous channel and (ii) the waveforms chosen adapt (or morph ) to match the opportunities for spectrum occupancy. Figure : Schematic of an interleaved bandsharing system using heteromorphic waveforms The advantage of the interleaved technique over the overlay method is that since underlay bandsharing is not time-constrained it can be used to ensure that the data transmission of the secondary user is time contiguous. Consequently, even when there are periods where there is no white spectrum available for the overlay, data can still be transmitted using underlay. This is an extremely powerful technique for maximising access to the spectrum and would appear to be the ultimate embodiment of a fully adaptive radio system for dynamic access to the licensed spectrum by bandsharing. It would almost certainly fall into the category of a cognitive radio. However, it requires far greater intelligence in order to provide optimum morphing to the prevailing RF environment. In such a system, a raft of different waveform types are available to the secondary user which can be chosen on the fly subject to its assessment of the RF environment and the actual transmission requirements of the user. Consequently the waveforms may include single carrier narrowband modulation, ultrawideband, bit loaded OFDM and direct-sequence spread-spectrum (or hybrids [83] - [86]) all chosen, as required, to minimise the level of interference to primary users yet provide all the transmission resources for the required data transmission [87] - [89].