[K3] AGC White Paper

[Richard Ferch-2]

> AGC adjusts the receiver gain, compared to the audio bits, rather slowly
and DOES NOT change the "dynamic range."

Exactly. AGC changes the receiver's gain between two points in time, not
between two signals arriving at the same time.

I believe changing the AGC settings does improve the "mush" situation in
CW, but the reasoning is more subtle than a simple picture based on dynamic
range alone.

Consider a situation where there are two CW signals, one at S8 sending
mostly dits, and another weaker one at S7 sending mostly dahs. If the AGC
threshold is well above the strength of either signal (or AGC is off), you
will be able to tell when a dit from the stronger signal ends even if a
simultaneous weaker dah is continuing, because of the change in signal
level when the dit stops. In other words, you will hear the dits from the
stronger signal riding in above the lower-volume-level dahs.

Now suppose the AGC threshold is below both signals, say at S6, and the
slope of the gain curve is flat (i.e. the SLP is set to its maximum value).
For now, assume a perfectly instantaneous AGC decay time (super-fast AGC).
Then regardless of whether one signal or both is/are currently "on", the
AGC will immediately adjust the gain to clamp the signal strength to the S6
level. In other words, even after the stronger dit finishes, the weaker but
longer-lasting dah will still be at the same perceived volume, and you
won't be able to tell when the dit ended. Result: you just hear the
combination of two signals at a constant volume, i.e. "mush". If there are
only the two signals, when both are "off" the gain will increase, raising
the perceived noise level, but that's not the "mush" problem, it's the
"noisy receiver" problem. If there are a lot of signals at or above the
threshold, then the "mush" would be more or less continuous.

In the real world, the AGC time constant will affect this. If the decay
time is long enough (as in slow AGC settings as used for SSB), you might
hear the audio volume drop immediately after the end of a strong dit and
then rise during a continuing weaker dah, whereas if the decay time is very
short, you might hear the brief drop in volume only as a blip in an
otherwise constant-level sound. But at first blush I would think this would
be a secondary effect compared to the main effect of adjusting the AGC
threshold to be above or below the level of the signals.

As far as I can see, none of the above applies to RTTY, which is nominally
a constant-amplitude signal. Whether the signal is above or below the AGC
threshold, the ratio of the instantaneous mark and space signals will be
the same (unless one of them is on the skirts of the filter bandpass). Even
if there are two signals on the same frequency, the differences in relative
strengths of mark and space from the two signals would not be affected by
AGC, although the overall amplitude might be.

I don't know enough about how RTTY decoders work to guess at whether a
constant-level signal (aggressive AGC) vs. a time-varying signal level (no
or weaker AGC) would have an effect on decoding, but the same reasoning
that applies to CW "mush" does not appear to me to be relevant.

73,
Rich VE3KI
______________________________________________________________
Elecraft mailing list
Home: http://mailman.qth.net/mailman/listinfo/elecraft
Help: http://mailman.qth.net/mmfaq.htm
Post: mailto:[hidden email]

This list hosted by: http://www.qsl.net
Please help support this email list: http://www.qsl.net/donate.html
Message delivered to [hidden email]

[Ed Muns]

Below is a thread from 7 March 2016 about AGC usage with RTTY decoders.
David Wicks, G3YYD, is the author of 2Tone and Kok Chen, W7AY, is the author
of CocoaModem.

Anecdotally, my experience after 250,000+ RTTY QSOs over the past 15 years
concurs that minimizing AGC action supports best decoder performance.  If my
ears, or widely varying signal levels, can't tolerate AGC Off, then I use
AGC Slow, SLP=0 and THR=14 or higher.

Note also the comments about receiver IF bandwidth of 500 Hz except in
extreme cases.  Even in big RTTY pileups such as I encounter sometimes in DX
locations, Again, I've anecdotally found that 500 Hz decodes better most of
the time.  I seldom go lower.  This also implies turning off the K3
Dual-Tone filter.

Both of these points (no, or minimal, AGC and moderate IF BW) are not
intuitive, especially for an experienced CW operator.

Ed W0YK
__________________________________________________________________

G3YYD, 0210:

Actually with RTTY the AGC setting should be slow.

The reason for this is the best decoders decode each tone separately and
make use of the signal amplitude and  measured noise over time.

They compare the individual tone amplitudes with their amplitude over about
one character time before and after the character being decoded. They then
combine the tones together before the final decision is made based on their
individual signal to noise ratio. Sudden changes to receiver gain will
provide less than optimum performance as it will alter the amplitude
relationship and noise over much less than 3 character times (about half a
second).

For those older decoders that use a FM demodulation system fast or slow AGC
makes no difference so set the AGC time constant as you would for SSB rag
chewing - slow.

As for bandwidth do not set it below 350Hz as Chen W7AY indicated earlier
this can cause distortion across the bandwidth by delaying some parts of a
RTTY signal more than others. This blurs one bit of the RTTY signal into the
adjacent bits. This is the signal causing QRM to itself. I personally tend
to use 500Hz on my K3 and only reduce to 350Hz in extremis. The filters in a
modern decoder are very narrow. 2Tone for instance uses a filter for each
tone that are just 45.45Hz wide and at 90Hz wide have more attenuation than
the receiver's dynamic range. Reducing RX bandwidth below 350Hz is for human
hearing limitations not that of the decoder.

73 David G3YYD
__________________________________________________________________

W7AY, 1015:

If you are willing to manually ride the RF/IF gain controls, "AGC off" is
best.

As David G3YYD has pointed out, you need the "gains" of the Mark and Space
tones to be perfectly equal.  Under poor SNR but good propagation
conditions, 0.5 dB of imbalance will cause noticeable harm in the error
rates.

Basically, you want the gains between the mark (M) and space (S) bits to be
constant.  The strength of the composite signal (M+S) need not be constant.

Together with proper filters (narrow enough to avoid QRM while adding no
intersymbol interference), slicing (deciding whether mark or space has
arrived) is an equally important aspect of FSK demodulator design.  You can
easily make the case that the slicer becomes more important when conditions
are poorer.

The slicer decides whether the mark signal or the space signal is greater at
each bit period.

Good demodulators take care of slicer imbalances by the use of "automatic
threshold correction" (ATC) circuits or software code.  You can also use FM
techniques to get around mark/space imbalance, but that creates more
problems that it solves -- that is why good demodulators nowadays use two
individual "AM" demodulators.

It is always best to present to the demodulator with a signal that has as
little possible tone imbalance so that the ATC has the least amount of work
to do.  

This way, you minimize the problems that the demodulator has to overcome.  

Thus, you would rather have AGC that does not keep the amplitude perfectly
constant, as long as the two tones have the same amplitudes.   Remember, the
key is to have no imbalance.  The two tones must fluctuate by the same
amount.

Good A/D converters (sound cards) provide dozens of dB worth of dynamic
range to handle fading.  Just keep remembering that RTTY demodulation
depends on SNR and not on signal strength.  Receiver requirements are very
different from voice or CW modes.

The ATC circuit has to work really, really hard (and fails often) when the
AGC is fast enough to be affected by the tone amplitudes fluctuating
independently.  The AGC time constant must therefore be much longer than a
bit period.  Even an AGC time constant that is around 176 ms (character
period of RTTY) already pose problems.  

Thus "AGC off" is the best, and if you are not willing to constantly ride
the RF gain control, the slowest AGC time constant possible is the next best
choice.

Use a A/D converter with good dynamic range, and let the demodulator
designers handle the rest for you instead of depending on the receiver
designers and their AGC circuits (few of them are designed with RTTY in
mind).

There really should be two channels from a receiver -- one that uses no AGC,
and is fed to the demodulator.  The other is a channel with AGC that goes to
the human ears.  That is how I embed an RTTY demodulator into my own SDR
program.  With floating point arithmetic, the channel that is fed to the
demodulator has practically unlimited dynamic range.

73
Chen, W7AY
________________________________________________________________

G3YYD, 1223:

I see Chen got into this one before me. Chen and I both design and make
available RTTY decoders. The internal workings of those decoders for best
performance requires that the AGC does not change the receiver gain abruptly
and it stays reasonably constant over several character times. Fast AGC will
cause problems with the automatic threshold "circuit" (software actually),w
which can be avoided by using slow AGC.

As Chen says manual gain control that is set and left is best of all but in
something like a RTTY contest not practical as the human also wants to hear
the signal at a reasonable level. So the compromise is slow AGC. Hang AGC
can also work well if the parameters are set correctly.

73 David G3YYD

______________________________________________________________
Elecraft mailing list
Home: http://mailman.qth.net/mailman/listinfo/elecraft
Help: http://mailman.qth.net/mmfaq.htm
Post: mailto:[hidden email]

This list hosted by: http://www.qsl.net
Please help support this email list: http://www.qsl.net/donate.html
Message delivered to [hidden email]

[Richard Ferch-2]

Aha! There's an explanation from the RTTY experts for the effects of AGC on
RTTY decoders.

Interesting that the recipes for best RTTY demodulation and for best
separation of multiple CW signals (the "mush" problem) appear to involve
similar AGC slope and threshold settings, although the reasons appear to be
different.

73,
Rich VE3KI


W0YK wrote:

Below is a thread from 7 March 2016 about AGC usage with RTTY decoders.
David Wicks, G3YYD, is the author of 2Tone and Kok Chen, W7AY, is the
author of CocoaModem.

...
______________________________________________________________
Elecraft mailing list
Home: http://mailman.qth.net/mailman/listinfo/elecraft
Help: http://mailman.qth.net/mmfaq.htm
Post: mailto:[hidden email]

This list hosted by: http://www.qsl.net
Please help support this email list: http://www.qsl.net/donate.html
Message delivered to [hidden email]

[Wes Stewart-2]

So if I slowly turn down the gain on one of your "pro audio" amplifiers and the
gain slope changes but remains a straight line that's distortion?

On 3/7/2017 4:11 PM, Jim Brown wrote:
> ...
>
> And I'll repeat my previous observation that a signal path with gain that
> varies with time or with the signal amplitude is NOT a linear device, and
> non-linearity => distortion.
>
> 73, Jim K9YC
______________________________________________________________
Elecraft mailing list
Home: http://mailman.qth.net/mailman/listinfo/elecraft
Help: http://mailman.qth.net/mmfaq.htm
Post: mailto:[hidden email]

This list hosted by: http://www.qsl.net
Please help support this email list: http://www.qsl.net/donate.html
Message delivered to [hidden email]