AVC with Russian Rod Pentodes

[kalee20]

I've been building a superhet radio with Russian rod pentodes this year, inspired by some of the posts by a couple of people on another Forum. My aim is a no-compromise, high-end design - and one of the issues I faced is provision for automatic volume or gain control.

I ended up with something which holds the output rock-steady between 1mV and 230mV RF input, and within 3db between 500μV and 500mV – that’s a 60db range.

Unlike pentodes with helically-wound grids (DL31, DL96, 1T4 etc) these little valves don't come in variable-μ flavours. I suspect it is theoretically possible (tapered or curved rods?) but I guess it's just too difficult to manufacture!

Anyway! To include AVC, two matters need to be dealt with. One is electronic control of gain; the other is deriving a control voltage or current.

For the first, some gain control is possible by varying grid bias. But, reducing gain by taking the grid nearer cut-off, reduces signal handling ability, just when we need it more. So this is a non-starter.

Varying filament voltage would probably work, but I don't like under-running filaments unless I can reduce other voltages and currents too, to avoid operating the valve in saturation-limited conditions.

Varying the screen voltage does work, and well. But it has the effect of reducing the cut-off voltage, thus reduces signal-handling capacity, though not so severely as biasing g1 more negative.

Applying negative voltage to the suppressor affects the balance between screen-grid current and anode-current, thus potentially reducing gain to anode. Unfortunately, it has the side-effect of reducing anode resistance, thus damping any anode tuned circuit the more, and reducing selectivity.

The strategy adopted was a combination of screen and suppressor control, and having comparatively low impedance tank circuits in the anode, damped by resistors to give the selectivity I wanted, so that the variable ra damping is swamped anyway. The consequential lower gain is overcome by having two IF valves (they're plentiful and cheap!), this also has the benefit that only moderate gain variation in each stage (10:1) is needed to give a large variation overall (100:1).

So, how to derive a control voltage? Several tens of volts swing is needed (+10 to +40V on screen-grid, and -20 to 0V on suppressors), so considerable amplification must be provided. And the screens take some hundreds of μA. I didn't want the AVC circuit loading the audio detector, and possibly causing distortion, so I used a buffer IF amplifier to isolate.

First attempt was to use an anode-bend detector, biased-off as a delay voltage. This worked, but the control voltage was modulation-dependent (note that the traditional double-diode triode topology suffers from this defect too). So I changed to a leaky-grid detector, filtering-out the audio with a (very) low-pass filter. This gives a control voltage of the wrong sense, so I then had to use another valve to invert it. But, it gives as a bonus additional DC amplification. I introduced the delay voltage at this point. With huge gain available, the circuit is capable of holding the output at a highly constant level even with big changes of input.

Eyebrows may be raised at the -60V bias line, but it's needed for other stages... And at very low power. It's easily derived from the heavily-screened and filtered switchmode converter which powers the whole thing from a 12V SLA battery.

I also fed some control voltage to the frequency changer (a single-balanced mixer using a pair of rod pentodes, with a third rod pentode as local oscillator driving suppressors in antiphase). Cutting gain in early stages prevents overloading in later stages – the final IF stage has only partial AVC applied to the screen. Of course, for lower-level signals it’s better to run early stages at full gain and then cut it in later stages because then you gut the noise too. It’s all a compromise!

Performance?

I ran a couple of tests: static audio output versus RF input, using a modulated RF generator and precision attenuator; and dynamic response by switching the RF input between two levels (6db apart) and examining how the output recovered to its controlled level.

See below for circuits, and results.

Comments invited!

[G.Castle]

That's a remarkable piece of design work Peter. And excellent results. There seems to be a flurry of interest in these little valves within the electronics homebrew community just recently, I had noticed that a few people were experimenting on the other forum, and a couple of friends have developed a pair of rather good small audio amplifiers, though nothing to rival the one you had built with some of the higher power handling Russian rod pentodes.

I think I may have to re-read this a few times to get my head around the principles you are employing here, its a lot more involved than anything I've seen before.

Well done!

(I understand from a mutual friend that there may be a operational video planned once finished?)

Regards,
Greg.

[radiomobile]

Have you considered the diode shunt system across an if transformer primary, the impedance, or perhaps I should say resistance, of the diode being controlled by the avc voltage ? This system was much used in transistor radios before the advent of ic's.

[kalee20]

Thanks Greg! Yes it's been a lengthy project, the valves may be little but it's by no means a minimalist design as I've used thirteen of them.

Radiomobile - yes I know what you mean. But shunting the primary of an IFT, while reducing gain, will also change the frequency response, something I didn't want to do (plus I wanted to keep this all-valve, no semiconductors at all). It's a workable system, as long as signals are at the millivolt level.

[G6Tanuki]

I'm fascinated by this: apart from the AGC [which you seem to be doing via a side-channel] I wonder what yiou're using as the detector?

Is it some kind of balanced circuit using two rod-pentodes? If so how are you feeding the carrier-insertion-oscillator to them?

At least if you're using a side-channel for AGC you're going to be less troubled with the issue of BFO/CIO leakage into the rest of the IF strip.

[kalee20]

Nope! I'm just using an infinite-impedance detector. No CIO, no balanced nothing! (Except for the frequency-changer which uses a single-balanced mixer).

Of course, the normal configuration of the infinite-impedance detector, looking like a cathode-follower, can't be used with directly-heated valves (unless a separate floating filament supply is used), but I got around that by just morphing the circuit around. (It's the main reason for needing the -60V bias rail).

[G.Castle]

Originally Posted by kalee20

some gain control is possible by varying grid bias. But, reducing gain by taking the grid nearer cut-off, reduces signal handling ability, just when we need it more. So this is a non-starter.


Comments invited!

This may be stupid question of the month but its been niggling me.

Doesn't this as you describe, also happened with the remote cut off type? I've never really played with gain control of valves, (by bias at any rate), I just thought that the variable pitch grid gives a wider range of voltage that is needed to control the gain, so easier to manipulate the controling circuitry, and that a valve near cut off would have poor large signal handling whatever its type?

Greg.

[kalee20]

You're right!

With conventional Automatic Gain Control, the valves are biased heavily negative to reduce gain. This takes them nearer cut-off.

The reason for wanting moderate or low gain is because the incoming signal is strong, so we have the situation that the valve is biased nearer cutoff just when we need it not to be, so that it can handle a strong signal without hitting cutoff and causing distortion.

However, a variable-mu valve does not cut-off so abruptly as a 'straight' valve, so the problem is not quite as severe as you might think. Add to this that in an IF amplifier, some types of distortion are tolerable or even irrelevant, because the tuned circuits restore the wave shape from a distorted sine-wave back to a relatively undistorted sine-wave. It turns out that second harmonic distortion is completely irrelevant, thought third harmonic and above do cause distortion of an amplitude-modulated signal.

[Radio Wrangler]

Good AGC is one of those problems without a real solution. You can't even get close to having all the desirable attributes at once. The usual approach is to design to take out the effects of signal strength variation on the wanted signal only and to just forget about the possibility of unwanted signals. This is OK-ish for broadcast receivers operating in their planned service areas. Adjacent signals (frequency-adjacent, that is) are kept distant (spatially) by spectrum management.

It is when you have to do a communications receiver and run into non-negotiable specs for adjacent channel rejection that you hit the real trouble. You have to chart signal level diagrams for wanted signals and unwanted signals at all sorts of offsets and levels in order to see where you run into non-linearity.

From this, you have to decide where to lose gain. It won't always be the same place or same pattern.

If this isn't bad enough, you need information to use to decide the gain settings. You can't use the detector at the end of the IF to detect the presence of out-of-channel signals because the IF filters them out, yet they may still be thrashing the earlier stages. You turn towards having multiple detectors spread through the receiver.

Now, you've got to keep the whole lot stable! - over a wide dynamic range.

IF agc characteristics tend towards a linear control voltage versus dB of gain characteristic (there is a reason for this). With a linear detector (like a straightforward rectifier) your loop gain varies proportionately to signal strength. Logarithmic detectors are one useful trick.

It's a bottomless subject.

DAvid

Good AGC is one of those problems without a real solution.

There is (at least) one, an ADC with sufficient dynamic range and low non linearity, both a bit below the natural noise level. Then chuck in a load of digital stuff afterwards. OK not AGC, no gain is changed in the analogue regime, seems the same to the end user. I have found that even the "cheap" SDRs such as SDRPlay do a very good job helped, no doubt, by their comprehensive front end filtering.

[Radio Wrangler]

Not even that works.

ADCs have limited dynamic range. Maybe you do some signal processing before them and if that includes anything active before the narrowest bandwidth ahead of the ADC, you have the same problem again.

If you apply the full incoming bandwidth to a very good ADC, then you have the full band taking up its peak voltage capability, then you reduce bandwidth with digital filtering and a DDC. But you have to truncate/round numbers to keep the DSP requirements affordable and you've got overload issues in terms of numerical overruns at different places in your progressive reduction in bandwidth.

You grow a few new freedoms in the DSP world, but some problems remain to haunt you. I've designed both fully analogue receivers and those with various levels of digital processing. I can testify that, when hard specs are involved, there is no golden bullet.

One of the DSP ones has to process a 4MHz bandwidth of noise simultaneously and cover 10MHz to 26GHz. It wasn't easy. It's proven to be OK as the industry standard for over a decade and somewhat over a quarter of a billion dollars worth of them are out in the field. Doing an AM receiver to meet the formal requirements for aircraft was actually a harder challenge.

David

[G.Castle]

Originally Posted by kalee20

You're right!

With conventional Automatic Gain Control, the valves are biased heavily negative to reduce gain. This takes them nearer cut-off.

The reason for wanting moderate or low gain is because the incoming signal is strong, so we have the situation that the valve is biased nearer cutoff just when we need it not to be, so that it can handle a strong signal without hitting cutoff and causing distortion.

However, a variable-mu valve does not cut-off so abruptly as a 'straight' valve, so the problem is not quite as severe as you might think. Add to this that in an IF amplifier, some types of distortion are tolerable or even irrelevant, because the tuned circuits restore the wave shape from a distorted sine-wave back to a relatively undistorted sine-wave. It turns out that second harmonic distortion is completely irrelevant, thought third harmonic and above do cause distortion of an amplitude-modulated signal.

Thanks for the explanation, appreciated.



Regards.
Greg.

Not even that works.

I did say an ADC with sufficient dynamic range and low non linearity. They don't exist yet, or I haven't heard of one yet anyway. One can dream...

[kalee20]

Thanks for comments, folks!

Strong adjacent-channel signals, at too early a stage for full selectivity to have been achieved and thus overloading these stages, is something I'd only considered at a very background level.

It's actually my first AVC system, in fact my first superhet (intended as a broadcast receiver) - I can see how many questions are being uncovered.

Originally Posted by merlinmaxwell

There is (at least) one, an ADC with sufficient dynamic range and low non linearity, both a bit below the natural noise level. Then chuck in a load of digital stuff afterwards... I have found that even the "cheap" SDRs such as SDRPlay do a very good job helped, no doubt, by their comprehensive front end filtering.

Originally Posted by Radio Wrangler

Not even that works.

ADCs have limited dynamic range.

Originally Posted by merlinmaxwell

I did say an ADC with sufficient dynamic range and low non linearity. They don't exist yet, or I haven't heard of one yet anyway. One can dream...

But if we could make such an ADC, we'd use the same technology to make an amplifier with sufficient dynamic range and low non-linearity, and do the same thing in analogue. David's concerns about 'out-of-band signals thrashing the earlier stages' causing overloading and intermodulation would disappear ...

Originally Posted by G.Castle

Thanks for the explanation, appreciated.

Hope it made sense!

But if we could make such an ADC, we'd use the same technology to make an amplifier with sufficient dynamic range and low non-linearity

Indeed, but I did say they don't exist yet!