Chapter V In this chapter we will explain the available transceiver architectures and will discus their specification, to find the best architecture to be used in wireless sensor network (WSN). WSNs are inherently low-power network with short rane communication capability. Most of WSNs are very dense networks and hence low cost of its nodes is an important requirement. RF transceivers are the most power consumin and hih cost part of a sensor node in WSNS. Consequently its low power and low cost requirements, demands new desin strateies. Prior to define a desin stratey it is necessary to define some parts of the networks communication protocol. Communication protocol and the radio link specifications insert limitations or offers freedoms on the transceiver characteristics. The most important characteristics of a receiver are: Dynamic rane Sensitivity inearity Noise fiure Phase noise Workin frequency and band width Channel selectivity Power consumption Spurious frequencies effect eakae effects Imae band reection Definition of dynamic rane depends on the specific application. This parameter is enerally a function of the transceiver parameters and the system requirement, like BER, VER, of false alarm rate, in detection systems. Dynamic rane, linearity and phase noise are not so important in low performance transceivers, like the ones is required in WSN applications. The important specifications of a transmitter are as follows: Power efficiency Power control capability Transmitted sinal power Workin frequency and band width Out-of-band emission (ACI 3, ACPR 4 ) Inter Modulation products (IMP) Transceiver requirements force the desiner toward specific class of transceiver. Then the transceiver architecture and its confiuration is selected or desined based on the specific application and the desiner s experience. Amon various specifications of the radio link, modulation scheme reatly influence the transceiver architecture. Simple modulations make possible to use very simple transceiver architectures, in expense of lower data rate, lower sinal quality and lower spectral efficiency. After selectin the transceiver architecture and confiuration, the required circuits and sub-blocks will be presented and finally the performance of the transceiver will be analyzed and optimized. Bit Error Rate Vector Error rate 3 Adacent Channel Interference 4 Adacent Channel Power Ratio 4
V. Transceiver Review Three architectures are classically used for transceivers in a radio link: Heterodyne, ow-if and Zero-IF or Direct Conversion (DCR and DCT ). Each of these architectures has some advantaes and some drawbacks, and hence desiner chooses the proper architecture dependin on the requirements of the application. Another transceiver technoloy, developed for short rane low power UWB applications is UWB impulse radio. In this technoloy very short, un-modulated pulses are directly transmitted from antenna [], [], [3]. This architecture is not suitable for simple narrow-band transceivers. NA-less receiver is another very simple architecture that may be used in wireless sensors networks as wakeup receiver [4], [5]. V.. Heterodyne Architecture Heterodyne architecture has been used in more than 98% of radio frequency applications until 995 [6]. Block diaram of heterodyne receiver has been shown in Fi. V-. Imae-reect filter is placed after NA and attenuates the imae sinal in the mixer input. Imae sinal has at f IF distance from the local oscillator frequency, in counter side of the massae sinal. To day this architecture is widely used in optical and electro-optical systems [7], astronomy and space science [8],hih frequency imain [9], and accurate and standard frequency measurement and spectral analysis [0], [], medical analysis [], and in mm-wave wireless communications [3], [4], [5]. This architecture has been noticed in modern CMOS technoloies for WSN applications [6]. The imae sinal problem is the bottleneck of heterodyne receiver, in sense of fully interated desin [7], [8]. Special techniques have been proposed this problem [9]. Imae sinal will be described in the next section. In eneral the imae sinal power can be even very hiher than the massae sinal [0]. So imae reection is the most problem of heterodyne receiver in radio communication applications. Choosin hiher IF frequency eases the imae reection, however the IF sinal processin and adacent channel reection will be more difficult [], []. In the next section, the low-if receiver that is a solution to reect the imae sinal is described. New architecture has been proposed for heterodyne receiver in [3] and claimed that the new architecture has many advantaes over direct-conversion receivers and relaxes the performance of the receiver buildin blocks and eases the overall system floor plannin. Heterodyne receiver can be desined as sinle or double IF stae. In classic heterodyne receiver channel selection is done by the RF local oscillator and hence the IF filter bandwidth is equal to the channel band width. In some other types the IF filter is wide and covers all the Fi. V-. Basic block diaram of a heterodyne receiver Direct Conversion Receiver Direct Conversion Transmitter 4
receiver operation band and channel selection is done in IF band [4]. This confiuration, named Wide IF heterodyne, has the benefit of more accurate channel selection, in expense of increased noise band width of the receiver. Main advantaes of heterodyne receiver are [6]: Hih selectivity (ood channel selectin in communication application and hih spectral resolution in spectroscopy and spectral measurements [8]) Hih sensitivity Hih dynamic rane (AGC easily be added to the IF amplifier) ess sensitive to DC offset of the mixer, spurious frequencies, hih frequency leakaes and even-order inter-modulation terms ess sensitivity to flicker noise Main drawbacks of heterodyne architectures are [7], [5], []: Imae reect filter problem in fully interated desin. In many applications, this filter should be implemented usin SAW filters or other technoloies that can not be interated in bulk CMOS technoloy. Inherently complicated (needs two VCO and some mixers) Hih DC power consumption IF filter problem in fully interated desin The NA must drive 50Ω load (the off-chip imae reect filter) and this adds to the power dissipation, ain and noise problems. V.. ow-if Architecture To benefit the advantaes of the heterodyne receiver, meanwhile makin it suitable for fully interated receiver and System-on-Chip interation, low-if receiver architectures were developed [5], [7]. In these receivers the imae sinal is suppressed usin complex sinal processin and without need for sharp passive filters, as in heterodyne receivers. This architecture is widely used in recent communication and wireless applications [6], [7],[8]. The low-if receiver reects the imae sinals, like an Imae-Reect Mixer (IRM), that its idea is based on the SSB sinal transmission theory, developed by Hartley in 98 [9], and Weaver in 956 [30]. Actually reection of the imae sinal is due to the fact that the desire sinal and the imae sinal appear in the IF band with different phase, as depicted in Fi. V- [3]. Fi. V-. Representation of imae reection in a low-if architecture, in frequency domain [3]. Automatic ain Control Sinle Side Band Transmitters 43
This receiver is implemented in two approaches [6], [9]. One is based on the Weaver method and the other is based on the Hartley method. The former also is known as phasin method. Two different implementation of low IF receiver architectures based on the Hartley method, have been shown in Fi. V-3 [3], [33]. Both of these use 90 0 phase shifter in the sinal path, that introduces the limitation in the sinal bandwidth, since the phase shifter is inherently narrow band. Weaver architecture has been shown in Fi. V-4 [5]. This architecture replaces the phase shifters in the sinal path, with the phase shifters in the local oscillator path and consequently does not suffers from the sinal bandwidth. Analysis of this architecture has been presented in [7]. Weaver architecture is suitable for diital implementation and has been widely used in recent years in CMOS technoloy [6], [7], [8].Mathematical interpretation of imae reection process in Weaver architecture has been iven in Appendix E. Poly-phase filters [34], are widely used in low-if receivers to improve the performance of low-if receiver [7], [35]. Imae reect performance of low-if receiver is hihly deraded in presence of the phase and ain imbalance in the mixer branches [6], [5]. In early CMOS technoloies the typical phase mismatch of 3 0 was restricted the imae reection to 6 db, but usin special techniques 46 db imae reection was possible [5]. In more recent technoloies 35 db imae reection is achievable with the standard low-if architectures, for below 0 GHz [9]. Some adaptive techniques have been used to improve the sinal path mismatch in a low- IF receiver [36], [37]. However these techniques are complicated and not suitable for low power applications, like WSN. A simple way to improve the imae reection of a low-if receiver is usin selective NA. Usin this method, in [38] 57dB imae reection at GHz has been reported and in [8] additional db improvement has been achieved. (a) (b) Fi. V-3. Two different implementation of Hartley low IF receiver: a) two phase-shifters with simple O and [33] b) one phase-shifter with quadrature mixer [3]. Fi. V-4. Architecture of Weaver low-if receiver [5]. 44
Weaver low-if architecture is a ood choice for WSN application, since it can be realized mainly in diital sections and hence yields a low-power receiver. Nevertheless, this is not the case for millimeter wave applications, in which hih value of mismatch occurs between the I and Q sinal passes. As an example, Razavi has reported.6db/6.5 0 mismatch in 60GHz band for 90nm CMOS technoloy [3]. Usin a simple analytic equation one can obtain an approximate value of imae reection []: ( ) ( ) δa cosθ δa IRR = 0lo ( δa) cosθ ( δa) (V-) In which A and θ are the ain and phase imbalance for two sinal paths. Insertin the above values in this equation we obtain 9.3dB imae reection that is very low for practical applications. On the other hand in mm-wave band some other crucial problems arise that corrupts the advantaes of the low-if architecture [3]. V..3 Direct Conversion Architecture Direct conversion receiver architecture (also reconized as zero-if or Homodyne architecture) was considered as early as 94, but as a crude receivers requirin only. In 947 homodyne receiver was used in full effect for carrier based telephony [39]. However superior performance of heterodyne receivers pushed it out in the radio communication systems. Emerin the RF-CMOS and trends toward fully interated transceiver, the zero-if architecture was considered aain. The first application of zero-if technique in diital receivers was in 980, in the FM pain receivers, firstly in ITT Standard Telecommunication aboratories, and then in NEC and Philips pain receivers [39], [6]. In recent years many researches have been turned to the direct conversion transceivers [40], [4], [4]. The architecture of zero-if receiver and transmitter has been shown in Fi. V-5. In the receiver, the RF sinal amplified by a NA then applied to a quadrature mixer. The local oscillator frequency is equal to the carrier frequency. Consequently the mixer translates the sinal in the RF band, directly into the base-band. In the transmitter a band pass filter can be used before the power amplifier, to suppress the excess interference and noise in the transmitted sinal [43]. If power control is required in a DCT, the power amplifier should handle variable output power. The most important advantae of the direct conversion receiver is solvin the imae sinal problem, and hence the imae reect filter []. On the other hand, in direct conversion receivers, the channel filterin takes place at base band, this has an advantae with respect to both interation as well as potential use in multi-standard and SDR applications [44]. With the desired channel modulated to base band, this enables the implementation of interated, hih-q filter architectures capable of providin sufficient reection of alternate channel enery before bein diitized. Because the carrier is directly modulated to base band, there exists the possibility of interatin prorammable base band sinal processin either in the form of prorammable filters or hih-dynamic rane ADCs followed by prorammable diital channel filters to address variable bandwidth and frequency response requirements associated with different standards. In comparison with low-if transceiver, base band sinal in a direct conversion transceiver has less band width than low-if architecture. Consequently the filterin in direct conversion architecture is less power consumin and ADC conversion can be done with half rate of low-if architecture and hence with lower DC power consumption [45]. Software Defined Radio 45
Fi. V-5. Architecture of direct conversion transceiver [45]. V..3. Problems with Direct Converter Receivers The most important problem is the DC offset. The DC offset may have static or dynamic nature [46], [47], [48], [6], []. There are three main sources for dynamic DC offset: First is the unsuppressed carrier in the received sinal that is directly converted to the DC sinal in the mixer output. Carrier sinal in the received sinal is due to the weak SSB formin and/or receiver back emission. The receiver back emission is due to the O sinal leakae to the receiver antenna and propaation to other receivers. This effect is more crucial in direct conversion receivers, because O frequency is exactly at the center of the NA and antenna band [49]. On the other hand the O leakae into the power amplifier in SSB transmitter causes the weak carrier suppression. This problem may be reduced usin heterodyne transmitters [50], however in heterodyne system the noise in the transmitted sinal due to the phase noise of O, that is added to the transmitted sinal (known as reciprocal mixin), is more than the direct conversion transmitter, in which only one O is used. The second contributor to the dynamic DC offset is the various leakae sinals, inside the receiver, as well as the unwanted RF sinals input from antenna [6]. Fi. V-6 shows the various leakae sinals. Self mixin of O sinal and the input sinal with their leakaes can be reduced with proper placin of mixer and NA in the receiver layout. The third contribution is the second order nonlinearity that leads to the detection of amplitude variations of the received sinal [5], [49], [4]. If the received sinal is stron desired sinal, this effect increases the BER and hence derades the dynamic rane. If the stron received sinal is not the desired sinal (Interferer sinal), then this effect derades the receiver sensitivity. An efficient way to overcome this problem is usin balanced circuits in the RF front-end. Fi. V-6. Self mixin of O sinal and the input sinal with their leakaes in a direct conversion receiver [6] 46
Static DC offset is mainly due to the transistor mismatch in the base band parts, after the mixer. Since the base band amplifiers in the direct conversion receiver have very hih ain, even very small mismatch leads to a noticeable DC offset in the sinal at the input of detector. In frequency domain, this offset appears in the middle of the down-converted sinal spectrum and may be larer than the sinal and much larer than the thermal and flicker noise. In the time domain the DC offset shifts the sinal constellation and increases the bit error rate. Most efficient way to reduce the DC offset is usin the modulation schemes by which the modulated sinal has very low enery in the frequency band around DC. By this way one can simply reect the DC offset by a notch filter in DC [5]. As an example, wide-band FSK is not spectrally efficient, but it has been widely used in low data rate systems, like paers [6]. FSK is a proper choice for wireless sensor networks, since it can be detected usin very simple detectors [53], [54]. For many applications, e.. diital cellular mobile communications, complicated spectrally-efficient modulations are needed. CDMA and WCDMA sinals have very small enery in vicinity of DC, but for GMSK modulation, used in GSM mobile system, the spectra has its peak at DC [5]. Spectra of some modulations have been shown in Fi. V-7. The other way to reduce the DC offset is usin adaptive compensations [55], [56], [57]. In TDMA systems, the DC offset can be measured in unused time slots and then the results can be used for DC offset compensation. However this method is useful only in the case of static DC offset. Burst-to-burst DC offset estimation, special feed back techniques and DC offset estimation are some techniques used for DC offset cancellation [58], [59], [60], [6]. In term of circuit desin, Even Harmonic Mixers (EHM) can be used to reduce the O leakaes [57], [6]. After solvin the DC offset, the flicker noise problem arises. Aain the easiest solution is usin sinal spectral shapin, so that the sinal enery distributed in the frequency band out of the reion in which the flicker noise is trouble. The other way is to use technoloies with low flicker noise corner. For example, the flicker noise corner for CMOS is about MHz and for Bipolar is about few kilo hertz [5], [63], []. Flicker noise reduction in a direct conversion receiver has been widely studied in recent years and mixers with flicker noise corner as low as few tens of kilo hertz have been reported [4], [64], [65]. One useful, but complicate technique to rune away the flicker noise is dynamic matchin mixer, in which usin two extra mixin (down-conversion and then up-conversion) the flicker noise spectral is separated from the sinal spectra [66]. Fi. V-7. Power spectral density for GMSK and CDMA modulations [5]. Frequency Shift Keyin 47
Carrier adustment and frequency dependent effects are other important problems in direct conversion receivers that can make the DC cancellation techniques ineffective. [67], [68], [6]. Even a small difference between the receiver O and the received sinal carrier frequencies can shift the sinal spectra, so that the sinal enery falls into the frequency bands in which DC offset or flicker noise is damain. One solution is transmittin pseudo-random data in some unused or dedicated time slots and averain the received sinal to correct the receiver O frequency [6]. V..3. Problems with Direct Converter Transmitters The most crucial problem in a direct conversion transmitter is VCO pullin [45], [68]. The most effective technique to reduce the unwanted effects on VCO in a direct conversion transmitter, is usin VCO in two-thirds of the O frequency. Then the VCO output is applied to a divide by two circuit and finally the VCO output and the divider output are mixed to enerate the desired O sinal [69]. However this technique produces unwanted terms in the output. In [45] another scheme has been proposed to obtain pure sinal in the synthesizer output. These techniques have been depicted in Fi. V-8. In hih performance transmitters, ow-if technique is used to overcome this problem [68]. The second problem with direct conversion transmitter is eneratin quadrature O sinal. In spite of heterodyne architecture, in which quadrature O sinals are required in low frequency (in IF stae), the direct conversion receiver and transmitter needs the quadrature O at RF frequency, in which the phase control is more complicated. Different techniques have been reported to enerate the quadrature O sinal. Choosin the proper one is trade of between hiher performance and lower power consumption [6], [70]. V..4 Transceiver Architecture for WSN Requirements of WSN, i.e. very low power and moderate performance has been forced the WSN transceiver desiners toward very simple transceiver structures. Fi. V-9 shows some the transceiver architectures developed for WSN applications. Simple receiver structure of Fi. V-9(a) detects the envelope of 96MHz RF sinal without convertin it to IF band [7]. In this fiure the transmitter is simply composed of an oscillator whose output is connected to the power amplifier input via a on/off switch (direct modulation). The transceiver shown in Fi. V-9(b) is another simple transceiver that uses super reenerative samplin oscillator to directly sample the.9ghz RF sinal. Both of the receivers in Fi. V-9 do not need to local oscillator. The transmitter in Fi. V-9(b) is a very low power BAW resonator oscillator and a simple sinle stae power amplifier. (a) (b) Fi. V-8. Frequency synthesize techniques to avoid VCO pullin in a direct conversion transceiver. The circuit in (a) produces unwanted harmonic in the output [69]. The circuit in (b) solves this problem in expense of more complexity [45]. Bulk Acoustic Wave 48
(a) (b) Fi. V-9. Simple transceiver structure developed for WSN applications, with direct detection of RF sinal usin (a) envelope detector [7] and (b) reenerative samplin [7] Rabaey et al. have reported power oscillator based transmitter in which the oscillator is used instead of power amplifier. They have used an accurate 90uW FBAR reference oscillator to luck the power oscillator by inection process (inection locked) [73]. Automatic Gain Control (AGC) is conventionally used in most of hih performance receivers to increase the receiver s dynamic rane. However AGC increases the complexity and power consumption and hence loarithmic amplifier or RF limiter is used as inherit AGC in low performance receivers. In [7] a loarithmic amplifier has been used and in [7] multi channel base band structure has been developed to increase the dynamic rane, mean while keep the receivers performance, without any power budet. Many other techniques may be used to simplify the transceiver structure. Direct connection of NA to antenna is a simple technique to simplify the receiver structure [74]. Envelope detectors have been used widely in WSN receivers, not only as amplitude detectors, but also as nonlinear low pass filter to discriminate the FSK sinals [75]. To address the problems related to VCO in direct conversion transceivers, VCO in half of the local oscillator frequency has been used in [76]. This technique also reduces the VCO power consumption. W. in Our Work Prior to desin RF transceiver, some specifications and parameters of the WSN should be iven. In our work, such data are not available and hence we will investiate the reported low power WSN transceivers to calculate some of the required parameters. A summary of Film (thin) Bulk Acoustic wave Resonator 49
reported WSN transmitters and receivers have been tabulated in Table V- and Table V-, respectively. V.. Radio ink Desin Since the radio link has not been defined for our work, we should determine and desin the items of radio like that are required in transceiver desin. A) Carrier and IF Frequency Reardin the reported WSN transceivers in Table V- and V-, all of the recently reported works have used carrier frequencies below 3GHz. However, hiher carrier frequencies have the main advantaes of hiher immunity and antenna interation possibility. These advantaes motivated us to try mm-wave band in our work. In recent years, mm-wave band has been considered as a candidate for low power short rane hih data rate TABE V- Summary of Reported WSN Transmitters Reference [73] [7] [75] [77] [78] [7] [79] Year 005 006 004 004 006 007 007 Technoloy 30nm CMOS 30nm CMOS 30nm CMOS 50nm CMOS 30nm CMOS 80nm CMOS 80nm CMOS Frequency.9.9.9 0.9.9 0.96 0.44 (GHz) RF Power -6 0.6-6 0.8 -.4 0 (dbm) * -. DC Power.8.6 3.6.3.35 3.8.58 (mw) **.8 9. Modulation OOK OOK OOK BFSK OOK OOK OOK BFSK Data Rate 5 50 40 0 330 000 40000 (kb/s) 56 Efficiency 5 3 6.5 9 46 NA NA (%) 8 * Radiated power, when TX is on ** Averae DC power, when TX is on TABE V- Summary of Reported WSN Receivers Reference [7] [75] [77] [76] [80] [74] [7] [8] [8] Year 005 004 004 007 007 007 007 006 005 Technolo y Receiver 80nm 30nm CMOS Noncoheren t 30nm CMOS Noncoheren t 50nm CMOS Noncoheren t Noncoheren t 30nm CMOS Noncoheren t 30nm CMOS Noncoheren t 80nm CMOS Noncoheren t 80nm CMOS Noncoheren t 80nm CMOS Noncoheren t Frequency.9.9 0.9.4.45. 0.96.4 0.433 (GHz) Sensitivity -00.5-78 -94-90 -90-53 -37-90 -50 (dbm) -65 DC Power 0.45 3.6.3 0.6 0.5 3. (mw) *.6 Modulation OOK OOK BFSK BFSK BFSK OOK OOK BFSK OOK BFSK Data Rate 5 40 0 00 NA 3000 000 00 000 (kb/s) BER e-3 NA NA e-3 e-3 NA e-3 e-3 NA * Averae DC power, when RX is on 50
communications, such as WAN and WPAN [83], [3], [84]. Hiher carrier frequency allows small and antenna, but increases the DC power drastically [85]. Obviously increasin the carrier frequency increases the path loss and hence more transmitter power is required. Mathematically, the path loss can be calculated as [8], [86], [87]: ( ) n 4π ( ) d path d = 0 lo n atten (V-) λ Where d is the distance of two nodes, λ is wave lenth and atten is the attenuation in the path. Althouh increased path loss forces the low power WSN desiners toward lower carrier frequencies, the network reliability and interference and ammin immunity is an important motivation for hiher carrier frequencies. Another important issue is the enery-per-bit value. Enery-per-bit is measure of comparin the enery efficiency of low power transceivers [7], [88], [89] and is calculated as: E b = Tb PDC (V-3) Where P DC is the averae DC power and T b is time duration of sinle bit. This equation implies that for a iven DC power, increasin data rate decreases the enery-per-bit, and this means more enery efficient transceiver. As mentioned, wider IF band is possible if hiher carrier frequency is used. Consequently, hiher data rate is achievable and reardin (V-3), this means that enery efficient transceiver is achievable with hiher carrier frequencies. However when data rate increased, the receiver bandwidth should be increased and reardin the receiver sensitivity equation [7]: Sensitivit y( dbm) = 74dBm / Hz 0lo( Bandwidth) SNR NF (V-4) the receiver sensitivity is reduced. This directly translates to the hiher transmitted power and consequently increasin DC power consumption. We have chosen the 30GHz carrier frequency in our desin to evaluate mm-wave band ability in WSNs. Selectin IF frequency is a compromise between the imae reect capability in one side, and the IF staes power budet and the detector performance, in the other side. Better imae sinal reection is achieved in hiher IF and better detection and more low power and flexible IF stae is obtained in lower IF frequency. We found GHz frequency as a ood compromise. B) Transmitter Power Reardin Table V-, transmitter s radiated power is about few milliwatts in reported WSNs. This ensures that the transmitter power consumption be in order of power consumption of other circuits of WSN node. In our work the power consumption of receiver blocks is predicted to be about 0mW. Consequently assumin about 5% efficiency for the transmitter, we chose 5mW as the radiated power of transmitter. Reardin Table V-, 5% efficiency in mm-wave seems difficult, but we have achieved it usin power oscillator transmitter. C) Modulation Scheme To overcome the band-width limitations, traditional cellular and wireless local area network (WAN) standards have rate emphasis on spectrally efficient modulation schemes, such as Gaussian Minimum-Shift Keyin (GMSK) or Quadrature Amplitude Modulation (QAM). In addition, coherent receivers are mandatory to increase the channel capacity for a iven bandwidth. However, these transceivers consume too much enery for sensor network applications. To address this problem, in 003 the IEEE approved the 80.5.4 standard for Wireless ocal Area Network Wireless Personal Area Network 5
low-power wireless personal area networks (WPANs). 80.5.4 supports both Binary Phase- Shift Keyin (BPSK) and Offset Quadrature Phase-Sshift Keyin (O-QPSK) modulation at a maximum data rate of 50 kb/s. Current 80.5.4 transceivers consume tens of milliwatts and have approximate enery per bit values of 00 nj/bit, which is less than cellular systems but still too hih for sensor network applications [7], [7]. To achieve very low enery-per-bit, very simple modulations, in conunction with noncoherent receiver structure have been used in WSNs (See Table V-). In a non-coherent receiver, in contrast with coherent receiver architecture, no oscillator is required for phase synchronization and the receiver can turn on quickly. Furthermore, when received power levels are lare, the power consumption can be dramatically decreased as little RF ain is required and no RF oscillation must be sustained. Two modulation schemes, i.e. On-Off Keyin (OOK) and Binary Frequency Shift Keyin (BFSK) have been widely used in WSN applications (See Table V- and V-). Fi. V-0 shows simple implementation of these modulations. Both of these modulations are constantenvelop and hence non-linear power amplifiers can be used. Actually theses modulations are not band-width efficient, but are enery efficient [74]. In comparison with OOK, FSK needs more complicated transceiver. FSK needs complicated VCO and in some cases two VCO is used [73], [90]. Detection of FSK is more complicated than OOK. OOK receiver enables the use of an envelope detection based receiver [7]. In [75] receiver has two branches and can operate as FSK or two OOK branch and noted that OOK is more preferable for dense WSN. In [87] PPM and OOK have been compared and usin analysis of battery life time, it has been deduced that for dense WSN OOK modulation is more enery efficient, but for spars WSN PPM is better choice. UWB sinalin was defined by FCC 3 in Feb. 00, is a ood candidate for low-power short rane communications and has some advantaes in WSN applications [9], []. However this technique needs complicated transceiver structure. We have chosen OOK in our desin and will analyze it in detail in the next sections. D) Receiver Sensitivity Receiver sensitivity is defined as the minimum receiver input power, by which the minimum performance of the communication system is achieved. Bit Error Rate (BER) is used as the performance measure in many communication systems and specially in WSN (See Table V- ). As deduce from Table V-, maximum BER value of e-3 has been accepted as standard value in WSN applications. So the obective in our work is to obtain BER less than e-3. Fi. V-0. Principle of OOK and BFSK modulations Pulse Position Modulation Ultra Wide Band 3 Federal Communication Commission 5
For a iven modulation scheme, BER is a function of SNR. Theoretical value of required SNR to achieve BER of e-3 with OOK modulation is 6dB [7]. In a dense WSN maximum rane (distance between two communicatin nodes) is about few tens of meters. For example, Rabaey et al. (Frontiers of WSN in Berkeley wireless research center) have desined their WSN node for less than 0 meters [73], [78]. Similarly, in [7] less than 0 meters rane has been considered. In [77] sensor node has been desined for 6 meters rane. In [80] 0-0 meters rane has been considered and in some other works the sensor nodes have been desined for 0 and 30 meters rane [76]. Reardin this suestion, we have considered the maximum nodes distance equal to 0 meters in our work. After determinin the transmitter radiation power and rane, the receiver sensitivity can be calculated usin Friis wave propaation equation [9]. It must be noted that accurate calculation of the path loss, specially for indoor applications is very complicated and Friis equation calculated the path loss for a point-to-point communication. However this equation can be used as a primary desin uideline in WSN applications [7]. Friis equation calculates the ratio of received power to transmitted power: Pr λ = GrGt (V-5) Pt 4πd Where P r and P t are received and transmitted powers, respectively, λ is wave lenth and d is the nodes distance. G r and G t are the receiver and transmitter antennas ain, respectively. Interation of antenna on chip is possible in mm-wave band and half- wavelenth dipole antenna is a ood choice for this purpose [93]. Maximum directivity of half-wavelenth dipole is.64 and hence for an antenna with 00% radiation efficiency, the ain of.5db is expected [94]. However, interated antenna in CMOS technoloies suffers from hih substrate loss and hence the antenna ain is considerably reduced. Half- wavelenth dipole antenna has been desined and measured in [93], in bulk CMOS and HR -SOI CMOS technoloies in K a band. In bulk CMOS, ain of -8dB has been measured at 36GHz and in SOI CMOS the measured ain is -db at 40 GHz. We consider half- wavelenth dipole in our desin and the antenna ain is assumed to be -8dB for both of receiver and transmitter. We assume -7dB ain for both of receiver and transmitter antennas at 30GHz. Now, for iven carrier frequency of 30GHz, node distance of 0 meter, transmitted power of 5dBm and antenna ain of -7dB, we can calculate the required receiver sensitivity from (V- 5). The calculated required sensitivity is equal to -90.5dBm. E) Data Rate and Receiver Band Width In the precedin section we calculated the minimum SNR and the receiver sensitivity equal to 6dB and -90.5dB, respectively. To calculate the receiver band width we use (V-4). For this purpose the receiver noise fiure must be determined. Based on our experience NF 8dB is achievable in our work. This is not strane, considerin the reported values of NF in mmwave band and 90nm CMOS technoloy [3], [4]. Now usin (V-4) the receiver band width is calculated: Sensitiviyt 74 SNR NF 0 BW = 0 = 890KHz (V-6) This is the maximum bandwidth of the receiver. We can choose the bandwidth less than 890KHz to ain better sensitivity or to relax the noise fiure requirement. The receiver data rate (bit-per-second) is chosen equal to the receiver bandwidth. Hih Resistivity 53
V.. Transceiver Architecture The specifications of the developed radio link have been listed in Table V-3. Based on the features of this radio link we propose the transceiver structure for our work. The transceiver architecture in our work has been shown in Fi. V-. To relax the isolation between transmitter and receiver, we have used TR switch, instead of duplexer. This is necessary, since in our desin receiver and transmitter are in the same frequency band. Remind that in networks for which the receivers communicate with a sinle base station, such as mobile phone cellular networks, TX and RX can be in different frequency. However in WSN applications all of the sensor nodes must have the ability to communicate with each other and hence the RX and TX must be at same frequency. All blocks of the receiver and transmitter can be switched to idle or active states to save the battery enery. In receiver branch, the Imae Reect Filter (IRF) has been placed prior to NA to reect the input sinals in the imae band. Even-harmonic mixer has been adopted for lowerin the VCO power consumption and to increase the VCO sinal quality [95], [96], [97], [98]. As will be explained later, in the even harmonic mixer the VCO frequency is half of the required O frequency in conventional mixer. This eliminates the frequency pullin due to the leakae sinals with frequency close to the VCO frequency [73]. In addition, VCO power consumption decreases drastically with decreasin the frequency [76]. Multi-slice IF amplifier and limiter, proposed in [7], has been used for increasin the receiver s dynamic rane and improvin the receiver performance, without any power budet. In each slice, the last IF amplifier acts as a limiter. At any time instance, only one slice is active, dependin on the received sinal strenth. IF slice (IF ain) selection is performed by the RSSI unit that measures the input sinal strenth [99]. The sinal amplitude is extracted by an envelope detector. Then a comparator is used for decide that the input data is or 0. Demodulation of OOK sinal is performed by envelope detection and simple comparison with a threshold voltae. Transmitter is composed of a simple power VCO. Althouh OOK modulations permit the use of non-linear hih efficiency class E power amplifiers, we encountered with some problems in drive stae of class E power amplifier in 30GHz band. Fi. V-. Proposed transceiver architecture with multi-slice IF amplifier and EHM in the receiver and power oscillator (PVCO) in the transmitter Voltae Controlled Oscillator Received Sinal Strenth Indicator 54
Table V-3 Features of the Desined Radio ink Carrier Frequency IF Frequency RX Sensitivity TX Power Channel Modulation Data Rate Band Width RX Noise Fiure 30GHz GHz -90.5dBm 5mW OOK 890KBps 890KHz 8dB V..3 The Receiver Architecture The receiver branch is composed of NA, VCO, even-harmonic mixer, multi-slice IF amplifier, envelope detector and comparator. The NA has been described in Chapter IV. Other blocks are described briefly here. V..3. VCO Block Frequency source of a coherent receiver is very complicated and is a power hunry part of receiver. However in WSN receivers in which receiver is non-coherent, a simple VCO is sufficient and as mentioned, even in the case of low RF frequency, VCO is not needed. Due to its low power, hih start-up reliability and ood tunin rane and differential output, crosscoupled oscillator is the most popular VCO circuit in CMOS RF transceivers [00], [0], [0]. In low power applications, complementary cross-coupled architecture is preferred over only-nmos cross-coupled architecture [0], [03], [04]. For equal bias current, voltae swin of complementary cross-coupled is twice of the only-nmos cross-coupled. Reardin to the esson phase noise equation [03], phase noise is inversely proportional with the square of the voltae swin. The main this advantae of cross-coupled VCO is its hih noise phase. Phase noise is the most important parameter of VCO for hih data rate communication, but is not important for low data rate non-coherent receivers [76], [03], [73]. In hih data rate receivers, phase noise causes itterin in the received pulses and this lowers the eye diaram openin in horizontal direction. If pulse duration is low, the eye openin reduction can cause noticeable increase of BER [05]. As mentioned, WSNs are low data rate systems and hence phase noise is not critical for them. V dd V dd V dd V tune V tune V tune V bias V bias (a) (b) (c) Fi. V-. Schematic view of NMOS cross-coupled (a), complementary cross-coupled (b) and current reuse cross-coupled (c) oscillators 55
Fi. V-3. Schematic view of Pierce oscillator (a), Miller oscillator (b) [07] and the oscillators proposed in our work: current reuse Pierce-like oscillator (c) and current reuse Miller-like oscillator (d) Fi. V-4. Small sinal model of Pierce-like (a) and Miller-like (b) oscillators Cross-coupled topoloy has been considered in WSN applications [03], [73]. Nevertheless other VCO circuits may be necessary in some cases. Due to low efficiency of power amplifiers in low power transmitters (See Table V-), the trend in WSN is to avoid them or use very simple power amplifiers. One solution is usin power oscillator in transmitter, by which the oscillator is connected directly to antenna [03], [73]. However when oscillator is connected to antenna, antenna load reduces the VCO quality factor and reduces the frequency stability. To overcome this problem, a simple ultra low power reference oscillator is required. For example, in [73] a simple 90uW FBAR oscillator with Pierce topoloy has been used for inection lockin the power oscillator. Some other simple oscillator circuits have been used in WSN applications [7], [75]. Tuned Input-Tuned Output (TITO) confiuration was reported in 008 in [0]. Current reuse is a common way to reduce power consumption in low power applications. Two current reuse oscillator circuits have been proposed for WSN applications. One circuit is obtained with some modification of complementary cross coupled oscillator [76], [79], denoted as current reuse cross-coupled, and the other is a Pierce oscillator [06]. We have examined three oscillator circuits in our work. First circuit is the above mentioned current reuse cross-coupled confiuration. This circuit has been shown in Fi. V- 56
, in comparison with NMOS and complementary cross-coupled oscillators. Two other circuits are modifications of Pierce and Miller oscillators [07], as shown in Fi. V-3 and we denote them as Pierce-like and Miller-like confiurations, respectively. So we excluded it from our investiation. To investiate the characteristics of three low power oscillators, i.e. current reuse crosscoupled, Pierce-like and Miller-like oscillators, and compare them we perform a simple analysis. A) Pierce-like Oscillator The small sinal equivalent of Pierce-like oscillator has been shown in Fi. V-4. Miller equivalent of ate drain feedback capacitance has been considered as a part of ate and drain nodal capacitances. Usin this fiure the loop ain is calculated as: V mn out mp = V ( C C ) Y (V-7) in Q where: ( C C ) Y = dn dp ( C Cd ) (V-8) ( C C ) Q Q is the inductor s quality factor: Q = (V-9) R And C and C d are equivalent input capacitances of two transistors at ate and drain nodes, respectively. Usin Miller approach and considerin that in oscillation the ate-drain voltae ain is equal to -, these capacitors are calculated as: C = Csp Csn ( Cdp Cdn ) (V-0) C = C C d dp dn n and p denotes for NMOS and PMOS transistors, respectively. Assumin that the equivalent capacitance in ate and drain nodes are equal, we can write: C = C C (V-) C = C Cd Substitutin (V-0) in (V-7) we rewrite the loop ain equation: V out mn mp = ( ) (V-) Vin dn dp C C C Q Q For a iven DC power consumption, the transistors small sinal model elements are determined and hence we have two desin parameters, i.e. and C. Oscillation occurs when the loop ain is unity. Consequently the oscillation conditions and oscillation frequency is obtained: dn dp C = Q mn mp (V-3) A > = dn dp ( ) dn dp dn dp Q Q Q 57
58 In practice there are many losses in the circuit, such as inductors and substrate losses that can be modeled as a shunt conductance at the drain node. Consequently, we rewrite the oscillation condition as: ( ) > = loss dp dn dp dn dp dn mp mn Q Q Q A (V-4) where loss is the conductance equivalent to the losses. If we assume ideal inductor, the equations reduce to: > = = loss dp dn mp mn A C f π (V-5) Obviously, more loss in the circuit corresponds to hiher m and hence more power consumption. B) Miller-like Oscillator In the case of Miller-like confiuration, usin Fi. V-4(b) we deduce: ( ) ( ) Y G C C C V V mp mn in out = (V-6) where: d dp dn C G C C G C G C Y = (V-7) And G i is equivalent to the inductor s loss, defined as: ( ) i i i i Q Q G = (V-8) Assumin that effective shunt inductance in ate and drain is equal, we can write: G C G G C G d = = (V-9) where is effective shunt inductance in ate and drain nodes and G represents the quality factor of the effective inductance and is calculated as: ( ) Q Q G = (V-0) C and C d are similar to the previous section, calculated in (V-0). Substitutin (V-0) and (V-9) in (V-7) we obtain: ( ) G C G C G G C Y dp dn =
And usin (V-6) we calculate the loop ain: V C( ) out mn mp = (V-) Vin ( dn dp ) C G G C G Oscillation occurs when the loop ain is unity and hence we obtain the oscillation conditions, after substitutin G from (V-0): Q Q C = dn dp ( ) ( ) Q Q C( mn ) (V-) mp A = > Q ( )( ) ( C) dn dp C C loss ( Q ) C) Current Reuse Cross-Coupled To analyze the current reuse cross-coupled oscillator, consider Fi. V-5. The oscillation occurs when the admittance seen from X-Y nodes is equal to the neated of C resonance circuit admittance. Reardin to Fi. V-5, from the small sinal model we can write: I I on op = = dsn dsp V V on op mn mp V V op on C C dn dp ( Von Vop ) ( V V ) op on (V-3) Subscribes n and p denotes NMOS and PMOS transistors, respectively. On the other hand, in the oscillator we have: Vosc = Vop Von (V-4) I = I on op So after some calculations we deduce: ( Cdn Cdp ) mn dsp Von = Vosc V op = dsn ( C C ) dsn dn mn mn dp dsp dsp dsn mp And the admittance to the X-Y nodes pair is obtained: Iop YXY = V = osc mp mp V ( C C )( ) ( ) ( ) dn dp dsp dsn mp mn dsp osc dsp dsn mp mp mp mn dsp C dp (V-5) (V-6) Oscillation occurs when Y XY is equal to the minus of the C resonance circuit. Consequently we obtain: Cdn( dsp mp ) Cdp( dsn mn ) dspdsn mpmn Q( Q ) = C (V-7) Q dsn mn After some manipulations we obtain the oscillation conditions: dsp mp ( ) 59
V op C sp mp V on dsp V op Y I op C dp I op X I on C dn I on V on C sn mn V op dsn V on (a) (b) Fi. V-5. (a) Equivalent circuit and (b) small sinal model of current reuse cross-coupled oscillator ( ) C ( ) Q Cdn dsp mp dp dsn mn C = ( Q ) dsn mn dsp mp (V-8) mnmp dspdsn Q > dsn mn dsp mp ( Q ) For compatibility with the two later analyses, we define the loop ain, considerin additional losses: ( Q )( mnmp dspdsn ) A = > (V-9) Q ( ) dsn mn dsp mp D) Comparin the Oscillators Usin the analytic equations derived in the previous section, we compare three oscillator circuits to choose the proper one for our applications. Reardin section V..., VCO frequency in our transceiver is half of the carrier frequency. Consequently, we compare three VCOs at 5GHz. Reardin the analytic equations, we have freedom in selectin the inductor value for all of the above oscillators. This is very useful, because the inductor has its individual optimization process, but each capacitance is easily by MIM capacitors. Based on our experience, we choose the inductance equal to 600pH for all three oscillators and the quality factor is obtained equal to 5. To obtain maximum swin, the NMOS and PMOS transistors are sized to achieve V dd / volts at the drain. With these conditions we have used our MOS transistor model for desinin three oscillators for various power consumptions. The loop ain of Pierce-like, Miller-like and current reused cross-coupled oscillators have been shown in Fi. V-6. The loop ains have been calculated usin (V-), (V-) and (V-9), respectively. Note that for all powers, the phase of loop ains are zero. This fiure has very important implications: For hih power applications, the current reused cross-coupled topoloy has excellent loop ain. Hih loop ain implies that the oscillator has very ood drive capability. However for low power 60
0 Pierce-ike Miller-ike CR Cross Coupled oop Gain 5 0 5 4 6 8 0 DC Power (mw) Fi. V-6. oop ain of Pierce-like, Miller-like and current reused cross-coupled oscillators at 5GHz, desined with various DC powers usin (V-), (V-) and (V-9), respectively. Normalized oop Gain 0.8 0.6 0.4 0. Pierce-ike Miller-ike CR-Cross Coupled 0 5 0 5 0 5 30 Frequency (GHz) Fi. V-7. oop ain of Pierce-like, Miller-like and current reused cross-coupled oscillators as a function of frequency, desined at 5GHz applications, less than mw in our case optimized Miller oscillator out performances the current reused cross-coupled topoloy. Consequently this oscillator is a ood choice for WSN applications. Optimized Miller oscillator has week performance, in comparison with two other oscillators. Fi. V-7 shows the loop ain of three oscillators, as a function of frequency. The oscillators have been desined for mw power consumption and 5GHz oscillation frequency. This fiure shows that the feedback network of Pierce-like and Miller-like oscillators have low-pass and hih-pass nature, respectively, as it is evident from their circuits. As we will show, hih-pass nature of Miller-like oscillator leads to semi-square sinal that is useful in power amplifier drive circuits. To compare the swin and phase noise of three oscillators, we have desined them usin the derived equations, for 5GHz and mw. Then the desined oscillators have been simulated in the STMicroelectronics CMOS 90nm desin kit. Fi. V-8 shows the phase noise of three oscillators. Miller-like oscillator has the best phase noise, -83dBc/Hz at 00KHz offset. As we mentioned, phase noise is not important in low data rate communications, especially in dense WSN applications. Waveforms of oscillators have been shown in Fi. V-9. From this fiure we deduce that Miller-like oscillator has semi-square sinal that is useful in drivin switch mode power amplifiers. 6
Phase Noise (dbc/hz) 0-0 -40-60 -80-00 Miller-ike Pierce-ike CR Cross Coupled -0 0 0 0 0 0 3 0 4 Frequency Offset (KHz) Fi. V-8. Phase noise of mw, 5GHz Pierce-like, Miller-like and current reused cross-coupled oscillators, simulated usin foundry desin kit and Spectre-RF simulator Miller-like Volts Volts 0.5 0 0 0 40 60 80 00 0 40 60 Pierce-like 0.5 0 ate voltae drain voltae ate voltae drain voltae 0 0 40 60 80 00 0 40 60 CR cross coupled Volts 0.5 0 PMOS drain NMOS drain 0 0 40 60 80 00 0 40 60 Time (psec) Fi. V-9. Wave forms of mw, 5GHz Pierce-like, Miller-like and current reused cross-coupled oscillators, simulated usin foundry desin kit and Spectre-RF simulator As we mentioned, the semi-square waveform is due to the hih pass nature of feedback network in Miller-like oscillator. Wave forms of Pierce-like and current reuse cross coupled oscillators are the same, but reardin Fi. V-6, the later has hiher drivin capability. In addition, current reuse cross-coupled oscillator is suitable for the cases in which VCO with differential outputs is required. In Miller-like confiuration, the inductors are rounded in one end and this makes possible use of line-type inductors. As a consequence of the above suestions, we have chosen current reuse cross coupled oscillator as local oscillator for the receiver and Miller-like oscillator as direct modulator and driver for class-e power amplifier in the transmitter. 6
V..3. Even Harmonic Mixer Sub Harmonic Mixers (SHM) offer an alternative solution to fundamental mixers, possessin some advantaes over fundamental mixers, in some applications. Unlike fundamental mixers, in which the VCO frequency is equal to the required O frequency, in n th order SHM, the VCO frequency is equal to /n of the O frequency. Due to some limitations, practical value of n is not reater that 4 [08], [84]. SHMs with even value of n are known as Even Harmonic Mixers (EHM). The leakae enerated by the coupled O sinal is the most important desin issue in direct conversion receivers. As explained in section V..3., O sinal leakae enters into the mixer and enerates DC components in the mixer output that derades the received sinal quality. This problem can be drastically reduced by usin EHMs [96], [09], [08]. The other advantae of SHM is reducin the pullin effect of the power amplifier output sinal. Power amplifier output is a hih power wide spectrum sinal and can leak into the VCO and reduce the VCO output quality by pullin its frequency [73]. In mm-wave applications, in addition of the above advantaes, usin SHMs allows use of low frequency VCO with better phase noise, hiher output power and lower DC power consumption [0], [84], [], [09]. Sub-harmonic technique can be applied for both of active and passive mixers. ike fundamental harmonic passive mixers, passive SHMs have lower DC power consumption and better noise performance, in expense of low conversion ain. In addition they are free of DC offset, an important issue for direct conversion receivers [09], []. In contrast to passive SHMs, active SHMs offers hih conversion, hiher linearity and hiher reverse isolation that makes them attractive for many applications, specially in mm-wave band [0], [96]. In frequencies well below mm-wave band, 4 th order SHM with doubly balanced Gilbert cell structure posses ood performance [08]. However this mixer needs 8-pahse O sinal that is very difficult to achieve in mm-wave band. Simple mixer circuits, such as ate-pumped [0] and sinle-balanced active CMOS mixer [96] are of more interest in mm-wave band. We have used the EHM circuit proposed in [96], with small difference. Conventional sinle balanced active CMOS mixer, the even harmonic mixer proposed in [96] and the similar one in our work have been depicted in Fi. V-0. In our desin IF filter has been mered in the mixer, to reduce the power consumption and increase the conversion ain. Here after we denote the EHM of Fi. V-0 as active CMOS EHM. In a conventional sinle balanced active CMOS mixer, the RF sinal is applied to the ate of tail transistor and the differential pair transistors are switched (in ideal case) on/off by O sinal. In contrast, in the active CMOS EHM the RF sinal is applied to the differential pair transistors and the O is applied to the tail transistors. The operation principles of both of mixers have been depicted in Fi. V-. To compare the performance of these mixers, we calculate their conversion ain in ideal operation condition. Assumin sinle-tone RF sinal, for conventional active CMOS mixer we can write: id = mvrf PT ( t) = mvrf cos( t) PT ( t) (V-30) id = mvrf PT ( t T / ) = mvrf cos( t) PT ( t T / ) where w is the carrier anular frequency, m is the tail transistor s trans-conductance and P T (t) is the pulse train with period T and unit amplitude. Usin Fourier series expansion of P T (t) and tackin the low frequency terms in the expansion of (V-30) we deduce: 63
Fi. V-0. Conventional active CMOS mixer, the even harmonic mixer proposed in [96] and the even harmonic mixer in our work id = mvrf cos( ( 0) t) π (V-3) id = mvrf cos( ( 0) t) π As an approximatin, we assume that m is proportional to the transistor s current, i.e. m( t) = ϕ id ( t) (V-3) φ is the proportionality coefficient. So from (V-3) we obtain: ϕpdc I IF = VRF (V-33) π Vdd where I IF is the amplitude of the IF component of drain current and V dd is drain supply voltae. In the case of active CMOS EHM we have: ϕpdc vrf id = PT ( t) Vdd (V-34) ϕpdc vrf id = PT ( t) Vdd And with similar approach we obtain: ϕpdc I IF = VRF (V-35) π Vdd That is exactly the same as the active CMOS mixer. This shows that with equal DC power, the even harmonic mixer has potentially the same conversion ain as the active CMOS mixer. To obtain an accurate analytic equation of active CMOS EHM, we deal with the realistic condition, in which the tail transistors are driven by a sinusoidal voltae, as shown in Fi. V-. Remindin that in short-channel MOS transistors, for lare ate-source voltae, the drain current is linear function of ate-source voltae, and reardin that the O sinal amplitude is well beyond the device threshold voltae, we can approximate the tail current as: µ CoxW it = ( Vs αvth ) Vs αvth (V-36) it = 0 Vs < αvth 64
Fi. V-. The operation principles of conventional active CMOS mixer and the active CMOS even harmonic mixer Where α is a constant equal to.. Accurate and approximated value of drain current has been shown in Fi. V. Usin this fiure we can calculate the DC component and first harmonic of the tail current: I I t0 t 4 = T 8 = T 4 = T ton o ton o µ C oxw V ton o ton µ CoxW µ CoxW ( V cos( t) V αv ) O O sin ( t ) 0 0 0 on ( V αv ) ( V cos( t) V αv ) cos( t) O 0 o o th th on ( t ) sin( 3 t ) o th dt t ( t ) 4 µ C oxw sin 0 on 0 on = ( ) sin 0 on VO Vo αvth T o 0 3 0 0 where T o and o are period and anular frequency of the VCO, respectively and: αv th Vo t = on arccos (V-38) o VO Now we can approximate the tail current: it It0 It cos( 0t ) (V-39) Note that hiher components of the drain current have no effect on the conversion ain, but they cause losses in the equivalent resistance between the supply and round. it ϕvrf id = It0 It cos( 0t ) cos( t) 4 (V-40) it ϕvrf id = It0 It cos( 0t ) cos( t) 4 Hih frequency components are filtered out and the IF current is obtained: ϕvrf iif ( t) = It cos( ( 0 ) t) (V-4) 4 The mixer normalized conversion ain (trans-conductance) is obtained as: ϕµ C ( ) ( ) ( ) ( ) oxw sin 0ton sin 3 0ton = sin 0t on CG N VO Vo αvth (V-4) To 0 3 0 0 0 dt (V-37) 65
I ta I t O O - V th V o Fi. V-. Representation of tail drive sinal (O sinal) of active CMOS even harmonic mixer On the other hand, power consumption of mixer is calculated usin (V-37): P = V I DC dd t0 ( t ) 8 µ C ( ) oxwvdd VO sin (V-43) 0 on = Vo αvth ton To 0 So we deduce: To PDC VO sin( 0ton ) Vo = αvth (V-44) 8V ddµ CoxWton ton0 t on is calculated usin this (V-38) and for iven P DC and VCO sinal frequency and amplitude, the VCO sinal offset (ate bias of tail transistors) is calculated. Then the normalized conversion ain is calculated usin (V-4). This makes possible to analytically calculate the optimum O drive sinal level and offset for the EHM. For example, normalized conversion ain of a 4GHz active EHM has been calculated usin (V-4) with different DC power consumption values and has plotted in Fi. V-3 as a function of O sinal level. This fiure shows that for each DC power, there is an optimum O sinal level. x 0-4 Normalized Conversion Gain 5 4 P =3mW DC 3 P =mw DC P =.5mW DC P =mw DC 0. 0.4 0.6 0.8 VCO Sinal Amplitude Fi. V-3. Normalized conversion ain of a 4GHz active EHM with different DC power consumptions, calculated usin (V-4) 66
In our work, the VCO frequency and RF sinal frequency are 4GHz and 30GHz, respectively. We have dedicated mw for mixer. The mixer was optimized usin our analysis results and usin optimization in Spectre-RF simulator and attached 90nm CMOS foundry desin kit. Periodic Steady State (PSS) analysis, in conunction with Periodic S-Parameters analysis was used for simulation of the desined mixer. S-parameters of the mixer have been shown in Fi. V-4. Both of input and output matchin are ood in the desired IF frequency. Conversion ain has been plotted in Fi. V-5. 4-dB conversion ain has been obtained at GHz. 3-dB band width is 80MHz and ain has ±0.5dB flatness in 60MHz bandwidth. Fi. V-6 shows the simulated sinle side band (SSB) and double side band (DSB) noise fiures. Fi. V-7 shows the power spectrum of the supply voltae and the mixer draws.mw DC power from V supply. Tail current of the mixer, obtained from simulation and from our analytic model has been shown in Fi. V-8. This fiure reveals the accuracy of our simple model. Performance of the desined mixer has been compared with the recently published mm-wave mixers in Table V-4. Based on our knowlede, as the table shows, our desin has superior performance for ultra low power applications. Fi. V-4. S-Parameters of the deined mixer, simulated in Spectre-RF simulator and CMOS 90nm foundry desin kit, with PSS and PSP analysis Fi. V-5. Conversion ain of the deined mixer, simulated in Spectre-RF simulator and CMOS 90nm foundry desin kit, with PSS and PSP analysis 67
Fi. V-6. Sinle Side Band (SSB) and Double Side Band (DSB) noise fiure of the deined mixer, simulated in Spectre-RF simulator and CMOS 90nm foundry desin kit, with PSS and PSP analysis Fi. V-7. Power spectrum of the supply voltae in the deined mixer, simulated in Spectre-RF simulator and CMOS 90nm foundry desin kit, with PSS and PSP analysis x 0-3 5 Tail Current (A) 0 3 4 5 6 7 Time (Sec) x 0 - Fi. V-8. Tail current of the desined mixer, obtained from simulation in Spectre-RF (a) and obtained from our simple analytic model (b) 68
TABE V-4 Comparison of Recently Published Ka Band Mixers Ref. Year Topoloy Technoloy Frequency (GHz) NF (DSB) (db) CG (db) DC Power (mw) [09] 008 D. Bal. Gilb 0.3um 4 0 3..8 * Cell EHM CMOS [3] 004 S. Bal. Gilb. 90nm 30 0.5 -.6 0 Cell CMOS [] 008 Gate GaAs 8~3 NA -3~-5 9.3 Pumped SHM [4] 007 Gilb. Cell 90nm 5~75 NA 3 93 CMOS [5] 007 Gilb. Cell 0.8um 8 5.5 40 CMOS This Work ** 008 Active EHM 90nm CMOS 30 5.8 4. * Plus RF pre-amp and IF buffer ** Simulation in Foundry desin kit V..3.3 IF and Base Band Circuits IF stae has been desined as multi-slice amplifier, the voltae ain of each channel is one decade hiher than the precedin channel. Reardin the received sinal strenth, proper channel and hence proper IF ain is chosen. This method eliminated the need for AGC and meanwhile preserves the advantaes of AGC over limiter or loarithmic amplifiers. At each instance in the on period of the receiver, only one channel is active and the others are in idle state with very small power consumption. Schematic of the multi channel IF amplifier has been shown in Fi. V-9. Each amplifier cell is a simple inverter-like class A amplifier with a resistor between ate and drain of transistors [74]. This resistor has the self-bias roll, as well as feed back effect to linearize the amplifier. The last amplifier cells in each slice has the roll of limiter amplifier. The multi-slice IF amplifier was desined and optimized usin Spectre-RF simulator and the 90nm CMOS foundry desin kit. Each CMOS stae draws about 00uA and 5nA in active and idle state, respectively. So maximum and minimum power consumption when receiver is on, is about 400uW and 00uW, respectively. Gain of the IF amplifier with different channel states have been shown in Fi. V-30. Noise effect of the IF amplifier may be considerable when Channel I is on, i.e. in the maximum IF ain. In this state the IF amplifier has the worst noise performance. To investiate the noise effect, we have simulated the output noise voltae spectral density, when Channel I is active, shown in Fi. V-3. The noise spectral density in around GHz is about 6µ V / Hz. Considerin about 00KHz bandwidth of receiver, the voltae noise standard deviation is obtained.68 mv. As will be explained, we have desined the receiver for 300mV IF sinal amplitude, prior to the detector and hence the noise of IF amplifier has no noticeable effect in the receiver performance. The envelope detector is after IF amplifier and detects the IF sinal amplitude. The envelope detector circuit schematic has been shown in Fi. V-3, that is a common-drain differential stae, biased close to the sub-threshold reion. This stae converts the IF voltae sinal to a rectified current sinal that is filtered by a RC filter that keeps the low frequency (base band OOK pulses) and rounds the hih frequency components. The voltae transfer curve of the envelope detector and its power consumption has been obtained usin simulation in Spectre-RF and has been depicted in Fi. V-33. The detector has linear response for the IF sinal amplitude reater than about 00mV. 69
Fi. V-9. Schematic of the multi channel IF amplifier, in CADENCE environment (a) (b) (c) Fi. V-30. Gain of IF amplifier channels when (a) channel I is active, (b) channel II is active, (c) channel III is active and (d) channel IV is active. (d) 70
Fi. V-3. The output noise voltae spectral density of the IF amplifier in maximum ain Fi. V-3. The envelope detector circuit schematic 7
800 00 Detected Voltae (mv) 660 460 75 50 5 Power Consumption (uw) 60 0 00 400 600 800 0 IF Sinal Amplitude (mv) Fi. V-33. voltae transfer curve and power consumption of the desined envelope detector as a function of the IF sinal amplitude, simulated usin Spectre-RF V..3.4 Receiver Performance Evaluation We evaluate the desined receiver performance in two ways, analysis and simulation. The performance measure in our work is bit error rate (BER), defined as the ration of the number of erroneous received data bits over total received bits, for enouh lare number of bits. A) Receiver Analysis Analytic equation of BER in OOK modulation in ideal condition, i.e. considerin all of the noises as Gaussian white noise, is obtained simply usin detection theory, as stated in many text books and literatures, e.. [6]. The resulted equation is as follows: Eb BER = Q (V-45) N 0 Where E b is the averae enery-per-bit and N 0 is the sinle-sided noise power spectral density. Q is the well-known Q-function, defined as: Q( x) = e dβ π x E b and N 0 are calculated as: E = P T b s b β (V-46) N0 = Pn BW P s is the sinal power, P n is the sinal power, T b is bit period and BW is the receiver bandwidth. So considerin that the receiver bandwidth is equal to inverse of bit period, from (V-45) we deduce: SNR BER = Q (V-47) where SNR is the sinal-to-noise ratio. Other useful representation of (V-45) is: Ps BER = Q (V-48) N0Tb P s is the received sinal power. When N 0 is the noise spectral density at the receiver input, this equation relates BER to the input sinal power and is used to fine the receiver sensitivity. 7
Sinle-sided noise spectral density in the input of a circuit with noise factor equal to F is calculated as: N0 = ( F ) kt V Hz (V-49) T is the ambient temperature and k is Boltzmann constant. In the case of our receiver, usin Frees equation we obtain: FMIXER N FNA kt G 0 = (V-50) NA We have calculated BER of our receiver, usin (V-48) and the results have been plotted in Fi. V-34. This fiure shows that with minimum acceptable BER equal to 0.00, the receiver sensitivity is equal to -89, db lower than the predicted value in primary desin. B) Receiver Simulation To more accurately evaluate the receiver performance, we have simulated it in MATAB, usin IF-Band modelin. Due to hih RF frequency, RF-samplin leads to enormous number of samples that can not be manipulated with ordinary PC facilities. So we have modeled the receiver system in IF band, i.e. at GHz. Each block of the receiver has been modelled behaviourally and parameters of different blocks have been obtained from accurate simulation of desined circuits in the foundry desin kit. Sinal source produces random 0 or data with uniform distribution and creates base band sinal equivalent to the random data. Then the base band sinal is filtered by the pulse shapin filter to limit the base band sinal spectrum. Pulse shapin filter is a Square Root Raised Cosine (SRRS) filter with 0. roll-off factor. Frequency response of NA and mixer has modelled usin proper filters, to capture the noise power reduction due to the limited bandwidth. IF amplifier has been modelled with ains obtained from simulation, and noise floor of 6µ V / Hz around GHz, as shown in Fi. V-3. The detector has been modelled based on the simulated characteristics of Fi. V-33 and proper low-pass filter. An example of the simulation results for a 00-bit data stream has been shown in Fi. V- 35. The received sinal level is -87dBm. Fi. V-36 shows the eye diaram of the received base-band sinal, correspondin to the 00-bit data stream. Eye diaram offers an intuitive view of the receiver performance, specially in the measurement time. In addition, usin eye diaram one can calculate the required input sinal power to achieve the ideal eye openin [05]. BER has been calculated from runnin the simulator for a 00,000-bit data stream and value of 0.00 has been obtained with -87dBm received sinal power. 0 - BER 0-0 -3 X: -89 Y: 0.00077-0 -5-0 -05-00 -95-90 Received Power (dbm) Fi. V-34. BER of our receiver as a function of the received sinal power 73
Voltae (V) Voltae (V) Base Band Sinal 0-0 x 0-3 0.5.5.5 3 x 0-6 NA Output 0 Voltae (V) Voltae (V) - 0 0.5.5.5 3 x 0-6 IFamp Output 0-0 0.5.5.5 3 x 0-6 0.5 Detected Sinal and Data 0 0 0.5.5.5 3 Time (Sec) x 0-6 Fi. V-35. An example of the simulation results of the receiver in MATAB, for a 00-bit data stream and - 87dBm received sinal level 0.8 Amplitude (V) 0.6 0.4 0. 0 0 500 000 500 000 Sinal Samples Fi. V-36. Eye diaram of received base band sinal, correspondin to a 00-bit data stream 74
C) Receiver Performance Summary Summary of the desined receiver and its performance has been listed in Table V-5. All of the receivers and transmitters in Table V- and Table V- are in the frequencies well below 30GHz and hence can not be compared with our work. Unfortunately we could not find any reported case of Ka band transceiver in CMOS technoloy. However some reported works are available in 60GHz band, but not for WSN application. We have tabulated the power consumption of these cases in Table V-6. The purpose is only to ive an insiht to the power consumption of conventional mm-wave transceivers. TABE V-5 Summary of Desined Receiver Parameters Carrier frequency (GHz) 30 Topoloy ACEH ** IF frequency (GHz) RF frequency (GHz) 30 Modulation OOK O frequency (GHz) 4 Sensitivity (dbm) -87 IF frequency (GHz) Bit Rate (Kb/Sec) 890 Conversion Gain (db) 4 DC Power (mw) 6.65 Mixer Noise Fiure (SSB) 7 Power Gain (db) 3.8 DC Power (mw). NA Noise Fiure (db) 3.6 S (db0) - DC Power (mw) 3 S (db0) -5 Topoloy CRCC * DC power (mw) <0.5 Center frequency (GHz) 4 Voltae Gain I 000 VCO Swin (mv) 450 IF Voltae Gain II 0 Phase noise @00KHz -83 amplifier Voltae Gain III 6 (dbc/hz) DC Power (mw). Voltae Gain IV 4.5 CRCC: Current Reused Cross Coupled ACEH: Active CMOS Even Harmonic TABE V-6 Comparison of our Receiver with Reported mm-wave Receivers Reference Year Receiver Topoloy Carrier DC power Technoloy Frequency (mw) [83] 006 Heterodyne (NAMixer) 60GHz 0.8 30nm CMOS [7] 007 Heterodyne 60GHz 77 30nm CMOS [4] 007 Heterodyne (NA Mixer O) 60GHz 80 90nm CMOS [8] 008 Heterodyne (IRRM with 30GHz IF, 60GHz 36 90nm CMOS without O) This Work 008 Heterodyne (EHM O) 30GHz 6.65 90nm CMOS V..4 The Transmitter Architecture We tried various reported nonlinear and switchin power amplifiers in 30GHz band, e.. cascode class C, driven directly by VCO [99], pseudo differential class B amplifier [9] and power VCO technique [73]. We also developed a novel desin approach for accurate desin of Class E power amplifiers [0]. Unfortunately, simulations in the foundry desin kit revealed that in 30GHz band forcin the NMOS transistor with enouh ate width, into on and off states, requires very hih power drive circuit and is not practical for a ultra-low power transmitter. Consequently we excluded the switch mode power amplifiers. On the other hand, other classes of power amplifiers have very low efficiency. Finally we found that the Voltae Controlled Power Oscillator (PVCO) is the best choice for our work. This technique was proposed by Rabaey et al. in the Berkeley Wireless Research Center, in 006 to avoid the DC power required for drive stae of power amplifier, in low power transmitter for WSN applications [73]. In this technique the VCO is desined with enouh power to directly feed the antenna, without any power amplifier. One problem with this technique is VCO frequency pullin due to the limited reverse isolation of antenna. To eliminate this problem, they used 75
an ultra-low power hih-q MEMS oscillator to increase the power oscillator stability by inection lockin. In the next enerations of their work, they replaced the power oscillator with lower power amplifier, directly matched to the antenna and achieved 46% transmitter efficiency at.9ghz [78]. Schematic view of the power VCO circuit in our work has been shown in Fi. V-37. The VCO has been desined as a conventional cross coupled CMOS topoloy. Usin N and P transistors leads hiher drive capability. OOK modulation is performed simply usin a switch MOS transistor at the tail of NMOS transistors. Simple calculation shows that to deliver 7mW RF power to 50Ω load, voltae swin of 830mV is required. To achieve this hih voltae swin, we have used hih voltae MOS transistors in the VCO. These transistors are provided by the STMicroelectronics 90nm CMOS technoloy and have.5v ate oxide breakdown voltae. Fi. V-38 shows the time domain wave forms of the transmitter, obtained from simulation in CADENCE usin Spectre-RF simulator and the attached foundry desin kit. From this fiure, the start-up delay of the oscillator is about.5nsec that is well beyond the requirements of a low data rate system. The oscillator output is very stable in the startup. Power consumption of the VCO and its phase noise have been shown in Fi. V-39. The VCO has noise phase of -85dBc/Hz at 00KHz offset and consumes 4mW power from.5v supply. The RF power delivered to the 50 load is reater than 6mW that corresponds with 5% power efficiency of the transmitter. Reardin Table V-, 5% power efficiency in 30GHz band is a superior result, if it proven with measurement. Assumin equal probability of transmitted and 0 bits, the averae power consumption of the transmitter is mw that is in ood areement with which we calculated in the radio link desin step of section V... Fi. V-37. Schematic view of the power VCO circuit in our work 76
Fi. V-38 Time domain wave forms of the transmitter, obtained from simulation in CADENCE usin Spectre- RF simulator and the attached foundry desin kit (a) Fi. V-39. (a) power consumption and (b) phase noise of the desined power VCO (b) 77