A discussion on the Automatic Gain Control (AGC) requirements of the SDR1000

Size: px

Start display at page:

Download "A discussion on the Automatic Gain Control (AGC) requirements of the SDR1000"

Marcia Dennis
6 years ago
Views:

1 A discussion on the Automatic Gain Control (AGC) requirements of the SDR1000 By Phil Harman VK6APH 1. AGC Characteristics 1.1 Static Performance The static characteristic of a receiver with and without an AGC system is shown in Figure 1 below: B C Output signal D F E A Input signal Figure 1 Receiver with or without an AGC The line A, B, C represents a system that has no AGC applied. The output increases linearly with the input signal until point B is reached, when some element in the signal chain overloads and becomes non-linear. Generally from point B to C the output signal is distorted and, unless the input signal is reduced, the system is unusable. The SDR1000 software can operate without AGC by selecting fixed in the AGC selection. If fixed is selected, the slope of the line A to B in the figure above - and the signal level at which the receiver overloads - is determined by the value of Fixed Gain selected under Setup>DSP> Options Increasing this value increases the slope of the line A to B and reduces the input signal level at which the signal distorts. Many direct conversion receivers do not use AGC. The advantages of not using AGC are simplicity and the purity of the received signal, in that weak received signals sound weak and strong signals sound loud. An incorrectly designed AGC system can introduce considerable distortion to an otherwise clean signal. The disadvantage of not having an AGC system is that the user needs to ride (rapidly reduce) a gain control to prevent their ears being blown off should a strong signal or a big noise pulse be received. Many low band DXers reduce their RF/IF gain to the point that the AGC is not operating on the (high) band noise

2 levels but will protect their ears in the event of the appearance of a sudden strong signal. More on this subject later. In my opinion, all AGC designers should be forced to spend some time using a direct conversion without an AGC as a first step in learning their art. The line A, D, E represents a system that has AGC applied. The slope A, D is greater than unity and indicates that the AGC has gain prior to the AGC detector. The transition from a linear to constant output at D is known as the AGC knee or threshold. From D to E the output level does not increase in response to an increase in input signal. How flat the section of the line D to E is depends on the overall AGC loop gain and is called the AGC Slope. Manufacturers often specify the performance of their AGC systems by stating the increase in output level in db for xdb increase in signal level above a certain threshold, e.g. less than 1dB increase in output level for an increase of 80dB above -100dBm (the AGC Knee). For some reason, receiver designers seem to think that the lower this level of increase is the better - more on this later. This level of increase is a measure of how tight the AGC system is. The alternate line D, F shows an AGC system whereby stronger signals are allowed to sound louder than weaker ones. The lines A,D, E and A, D, F also illustrate two common errors in the design of AGC systems. These errors are prevalent in many of the popular brands of Japanese transceivers and are: 1. The AGC knee is set at too low an input level. This has the effect of applying AGC either to band noise or, at an extreme case, the front-end noise of the receiver. Such a radio is very tiring to listen to for long periods, particularly when used at VHF/UHF where white noise appears for long periods. Such receivers sound flat since the AGC Knee is so low that all signals have the some volume. 2. Above the AGC knee the AGC characteristics are virtually flat (i.e. very tight AGC). Weak signals and strong signals all sound the same and again the receiver sounds flat. For many users, a pleasing alternative can be achieved by making strong signals sound louder than weak ones. A slope of between 6 and 10dB is usually sufficient to achieve this effect. A slope greater than 10dB introduces the potential for very strong noise pulses to sound uncomfortable, particularly when the user wears headphones. I am the first to admit that there are many receivers in daily use that have the characteristics of 1 and 2 above and are perfectly acceptable to their users. This is most likely due to the fact that the user has never used a receiver that allows these knee and slope values to be adjusted. In my experience, it is not possible to pre-set either the knee or slope of the AGC curve and obtain a performance that will satisfy all users. Experienced home brewers set these values to suit their individual tastes and operating skills, but more importantly, to match the band conditions in place at any particular instance. It s difficult to imagine how one AGC setting is going to suit working weak DX through summer QRN on 160m and moonbounce on the VHF+ bands. The solution to these differing requirements is to expose the AGC settings to the user. This is simple to do in conventional linear AGC systems and even simpler in digital systems. Few receivers actually make these controls available, but the software design of the SDR1000 means they can be implemented in an almost ideal manner. 1.2 Dynamic Performance Fundamentally an AGC system is a closed loop control system, hence all the issues related to such systems need to be considered in their design. Whilst the previous static issues are important, the dynamic issues are equally important since getting these wrong often results in a clean input signal becoming distorted.

3 The most common AGC design errors are overshoot, delay and instability. AGC overshoot is discussed later. AGC delay occurs when the AGC voltage is applied too late, with the result that the leading edge of a signal is louder (and hence potentially distorted) than optimum. AGC instability can occur due to the AGC loop oscillating in response to an input signal. These oscillations then amplitude modulate the incoming signal, causing distortion. Prior to v1.4.* of digitalagc.c, the SDR1000 suffered from the problem of AGC delay when the peak value of the input signal appeared within the first 1mS of the data block. In order to evaluate the dynamic performance of an AGC system, we stimulate it with typical input signals and determine the desired outputs. Typical input and desired outputs are shown in the following figures: Noise Burst B Noise Burst Signal Input Envelope A C D E F Pause in speech G H I Fade J Band noise Signal Output Envelope K L M N O P Q R S T U V V Short tc detector AGC Voltage W X Y Z Long tc detector

4 Note: The above diagrams are a variant of those in the Plessey Semiconductors SL600 Series Applications Manual A discussion of the performance of the above AGC system follows. The static performance of the AGC system has been set so that the AGC knee is well above the ambient band noise level and the slope of the AGC curve is non-zero. The AGC system consists of two detectors with different time constants, one short and the other long. In practice, two detectors are needed. Many analogue receivers use a single detector with a fast attack and slow decay - we will see later why this is not an optimum solution. Implementing dual time constant detectors in software is a trivial exercise, so unfortunately this short cut is also starting to appear in digital AGC systems. The actual AGC voltage generated at any instant is the larger of these two detector outputs. The lower diagram assumes that an increase in AGC voltage will result in a decrease in system gain. At A, the input to the system is band noise. Since the AGC knee is above the band noise, AGC action does not take place and the output is the demodulated band noise K and there is no AGC voltage generated at W. At B, a large narrow noise pulse is presented. The short time constant detector produces a large voltage X and limits the intensity of the resulting output to L. At C, the input reverts to band noise again. Since the dynamics of the AGC system is still determined by the short time constant detector, the output rapidly returns to its previous level. Note the importance of using a short time constant to process noise pulses. Had we used a long time constant here, then the AGC would not have had time to react to the narrow noise pulse and a large pulse would have been present at the output. Conversely, if the pulse was long enough to cause the long time constant detector to operate (or we had used a single fast attack/slow decay detector) then the output would have been muted whilst the long time constant decayed. At D, a speech signal is presented and the short time constant detector produces an AGC voltage at Z. This results in the output level N. Since the input signal is present for some time, the long time constant detector has sufficient time to charge and takes over the AGC voltage at 1. At E, the input signal level increases at such a rate that it can be tracked by the long time constant detector. Since the slope of the AGC system is not zero then there is some increase in the output level at P. The AGC voltage tracks as expected at 2. From F to G the envelope of the input stays constant and the system is stable. At G, a pause in the speech takes place. The output of the short time constant detector rapidly falls to zero at 3, whilst the longer time constant detector decays slowly so as to maintain the AGC control voltage during the pause. At this point two options are available. We can either maintain the AGC control voltage at the pre-pause level or allow it to decay. Using the former, then a so-called hang AGC system is implemented. This is particularly effective when the input signal is from a strong SSB signal. During a pause in speech, the AGC voltage hangs for a period such that the output is at a very low noise level or, if the signal is strong enough, virtually zero. This is point R in the previous diagrams. My preference is to use full hang AGC for signals above a user-selectable signal level. Signals below that level use the exponential decay of the longer time constant detector to control the AGC signal.

5 Note that at the end of the hang period, the AGC volts drop to a low level very rapidly, so that the AGC gain is restored quickly. This is to ensure that should the pause be overly long, then the receiver is ready to adjust its gain ready for the next transmission which may be significantly weaker than the previous, e.g. in SSB net-type operation. The implementation of controls to set the signal threshold above which hang AGC is enabled, and the hang time, is straightforward in both analogue and digital systems. At H, the pause is over and the signal returns at its previous level - the output level at S and AGC voltage at 4 are consistent with this event. At H a short noise pulse is presented above the level of the existing signal. The short time constant detector produces a suitable control voltage at 5 and, due to the non-zero slope of the AGC curve, a slight increase in output is present for the duration of the pulse. Note that the long time constant detector is not able to respond to the narrow noise pulse, but holds the previous AGC level so that it is available again once the pulse has been removed. At I, the input signal starts a slow fade such that its rate can be tracked by the long time constant detector. The output level at T falls slightly due to the non-zero slope of the AGC curve and the AGC volts track as expected at 6. The input signal becomes stable again for a while and then is removed at J. Prior to being removed, the input signal has dropped below the threshold at which the hang AGC is activated. Hence the AGC long time constant detector decays exponentially and the output level increases exponentially back to the level of the band noise. One of the most common errors in AGC design results in what is known as AGC overshoot. This occurs when the time constant of the AGC system is not fast enough to track the envelope of the incoming signal. The sequence is; the signal appears at the input and, due to the delay in the AGC system, immediately appears at the output. The AGC then reacts and provides a corrective signal, frequently too large in value. The output signal then drops but shortly after has to increase again, since the AGC signal has overshot the correct signal. Most analogue AGC systems have this to some degree - the fast AGC setting on my Yaesu FT847 is perhaps the worst example of this I have ever encountered! In most cases AGC overshoot is a simple problem to overcome in analogue systems, assuming there are not too many delays in the AGC loop caused by multiple narrow band IF filters. With fully digital AGC, system overshoot should never be encountered. With regard to many of the receiver problems encountered in the current crop of commercially produced HF//VHF/UHF transceivers, I suspect that many of the designers do not take their designs home and actually use them to receive signals under real conditions! Most of the AGC shortcomings of the receivers of today are obvious to anyone who has spent some time listening to a receiver from the 1960/70s golden age of amateur radio such as the Drake R4C, which has a wonderful AGC system for SSB. Before looking at digital AGC systems, and in particular the AGC in the SDR1000, one final point regarding the decay characteristics of AGC systems needs to be made. Since virtually all analogue AGC systems rely on the charge or discharge of a capacitor to provide the necessary time constants, inevitably such decays are exponential. Thus when a large signal is removed from the input of a receiver, the output level will return to the noise floor and this noise will increase in an exponential manner. The sound that a receiver makes when recovering from a strong signal is often talked about by old-tyme RF design engineers as how well the AGC system breathes. It s not difficult to see how this terminology came to be used since the sound a well-designed AGC system makes when recovering from a strong signal is not dissimilar to one of us breathing out. It is certainly possible to develop an analogue AGC system with a linear decay (by discharging the timing capacitor through a constant current source). However, it is also simpler to code a linear decay when using

6 software rather than an exponential one, so there is a danger that software engineers are going to take this approach! If we build an AGC system so that we can switch between an exponential and a linear decay, then we will see a marked difference in how the two systems sound. By setting the exponential decay at five time constants to the same point on the linear decay, then both systems should reach zero AGC volts at the same time. However, the two systems will actually sound completely different. The exponential delay appears to recover in a more audibly pleasing fashion, with the band noise returning progressively. With the linear decay the band noise appears with unnecessary haste towards the end of the time constant. I suspect that the difference is due to the non-linear level response of the human auditory system, but further work is necessary. Having gone to the trouble of implementing a switched decay characteristic in software, I now find that I never use the linear decay code. 2. AGC in the SDR1000 Please note that whilst these comments relate specifically to the AGC software currently used by the SDR1000, they apply to the design of digital AGC systems in general. I say currently since the software used by the SDR1000 changes frequently and the AGC shortcomings that are present in the version available today may soon be eliminated. I have incorporated all these changes in my own modifications to the last VB version and the screen shots that follow are from this work. The current SDR1000 AGC system operates on a block of audio samples from the associated PC sound card. Whilst the sound card sample rate is adjustable, the following analysis assumes a sample rate of 48kHz and a block length of bit samples. The nature of the SDR1000 architecture means that there is a delay between the signal being applied to the antenna and being heard by the user. Such delays are usually insignificant in analogue receivers but can reach quite noticeable - and annoying - levels in digital radios. The delay in the SDR1000 software is noticeable but only really significant if fast QSK systems like PACTOR or full break-in CW are to be used. In practice, some delay is highly desirable since this enables AGC actions to be fully applied to a block of audio samples before they are passed to the user. This is a very useful byproduct of a DSP-based radio (that does all its AGC digitally) and would be significantly more difficult to implement in an analogue receiver. In the SDR1000 AGC implementation prior to v1.4.* the AGC is implemented by first looking for the highest signal sample in the current block. This value is then compared against a pre-set value. If the highest sample value is different to the pre-set value, a multiplier value is calculated and applied to every sample in the block so as to make the highest level equal to the preset level. In fact, the new values are not applied instantaneously to all sample values but are initially implemented in a ramped manner for 1mS this provides a fast attack to the AGC system. More details of the software that implements this system are available on the SDR1000 web site The period over which the multiplication factors are applied is varied, depending on whether the input signal is increasing or decreasing, hence providing different AGC time constants. Additionally, the decay time can be selected from Fixed, Long, Slow, Med and Fast by the user.

7 There are a number of problems with implementing an AGC system in this manner. Firstly, since the multiplication factor is applied to all samples a single large noise pulse will capture the AGC and prolong the impact of the pulse. If a long AGC time constant is selected, the effect can be quite decremental to the performance of the receiver. Secondly, since the multiplication factor is not applied immediately but in steps over 1mS, if a large signal appears within the first 1mS of the data block the leading edge may not be sufficiently attenuated. As mentioned previously, this problem has been addressed in v1.4.* by using a circular buffer which enables samples from the previous block to be considered when determining the multiplication factor. It should be pointed out that prior to addressing this problem, the audio quality of the SDR1000 was already outstanding, mostly due to the use of phasing techniques, very low distortion hardware and high performance DSP software. With this problem corrected the audio produced by the SDR1000 not only sounds better, but is measurably better than any previous profession or amateur receiver that I have tested in over 38 years of RF design experience. Block Diagram Fast Time Constant Detector AGC Slope AGC Knee/ IF Gain Slow Time Constant Detector IF in The following screen shot shows some of the controls I implemented in the VB software to enable the AGC characteristics to be adjusted.

The AGC Hang Threshold slider sets the level at which the hang AGC system is implemented. For signals below this threshold, the AGC time constant reverts to Fast.

8 The AGC Hang Threshold slider sets the level at which the hang AGC system is implemented. For signals below this threshold, the AGC time constant reverts to Fast. As an aid to setting the threshold, a LED is provided on the Control panel that lights whenever a signal exceeds the set level. The AGC Slope slider sets the slope of the AGC curve once the input signal is above the AGC threshold. The remaining control, AGC Threshold/IF Gain, sets the AGC Threshold level. The green LED adjacent to the control lights up if the input signal has exceeded the level set. This is useful in that it enables the threshold to be set just above the band noise. A further control selects the AGC time constants of Manual, Fast, Medium, Slow and Hang, as well as selecting Off. In Manual mode, additional sliders (not shown), allow the user to individually adjust attack and decay times. Conclusions Current and future AGC systems can benefit greatly by a careful application of the rapidly disappearing old-tyme art of AGC design. In particular, the following need to be considered: Provide all the necessary controls to enable the user to adjust the AGC characteristics to suit their individual skills, operating habits and prevailing band conditions. This may be difficult for analogue radios, given the additional panel space required. For PC-based software receivers, there are no excuses for not providing this degree of flexibility. In particular the following controls should be provided: AGC Off, Manual, Fast, Medium, Slow and Hang AGC settings. An exponential decay for Manual, Fast, Medium and Slow settings. Sliders to set the attack and decay times in Manual mode. Hang threshold control. An LED to indicate the Hang threshold has been exceeded. Hang time control. AGC knee level control becomes IF Gain when AGC is Off. An LED to indicate that AGC action is taking place. AGC slope (0 to 10dB) control. The use of two AGC detectors; one with a short time constant and the other with a long time constant. The instantaneous AGC voltage is the larger of the two detector outputs. A short noise pulse should not trigger the long time constant, nor prove distressing to the operator. The receiver should recover immediately to full gain once the pulse has been removed. ends

HF Receivers, Part 2

HF Receivers, Part 2 Superhet building blocks: AM, SSB/CW, FM receivers Adam Farson VA7OJ View an excellent tutorial on receivers NSARC HF Operators HF Receivers 2 1 The RF Amplifier (Preamp)! Typical