Instantaneous (analogue) compression of speech signals

[Ken Smith]

I'm puzzled by "RF/IF" clipping. How does that work to improve the
demodulated audio?

In a single side band reciever:

If you've added 1MHz to all the frequencies, and clipped the signal in
that form, the results are quite different than clipping the detected
audio. For a single input frequency, all the distortion products end up
above 2MHz and thus get removed by the filtering.

The distortion ends up being heavy on the IM distortion effects and light
on harmonics. For some reason, this seems to be easier on the hear.

 

[John Woodgate]

I read in sci.electronics.design that Lasse Langwadt Christensen
'Instantaneous (analogue) compression of speech signals', on Fri, 7 Jan
2005:

I have a pair of speakers that use a lamp it to protect the tweeters,
but
I'd say its more of a power limiter than a voltage limiter

Loudspeakers are voltage-operated, in the sense that they give the
designed frequency response with constant-voltage input, not constant
power input. In any case, a series element can't divert any current; it
can only reduce current as a consequence of developing a voltage across
itself.

and wouldn't it violate your requirement of not having a time constant
to
worry about ?

Yes. A very long, thin filament in a gas-filled lamp might have a
thermal time-constant for fairly small temperature rise low enough not
to matter, but that's very low; less than 10 ms.

 

[John Woodgate]

I read in sci.electronics.design that Ken Smith
about 'Instantaneous (analogue) compression of speech signals', on Fri,
7 Jan 2005:

The distortion ends up being heavy on the IM distortion effects and
light on harmonics. For some reason, this seems to be easier on the
hear.

Not in my experience, but that is with systems having wider bandwidth.
One would expect IM to sound worse, because the IM products are not even
vaguely harmonically-related to the fundamentals.

 

[John Woodgate]

I read in sci.electronics.design that Ken Smith
about '"all pass" thought about (analogue) compression', on Fri, 7 Jan
2005:

That sounded clear to me, but I already know what I was thinking.

It sounds clear to me, as well, and I expect it would work, at least to
some extent.

 

[gwhite]

In a single side band reciever:

It is actually done in the *transmitter* as a purposeful processing technique.
This is because clipping at baseband tends to "square up" waveforms. When
passing the signal through a hilbert transformer (as is needed for the SSB "Q"
channel), the transformer increases the peak power requirements. For example, a
hilbert transformed square wave has infinite peaks. So at least part of the
benefit of RF/IF clipping is the reduction of peak power required. The other
benefit is a complete absence of harmonic distortion products due to the BP
filtering.

FM and AM don't per se "need" RF/IF clipping, but could use it effectively too.
(In that case the processed audio will be "re-basebanded" before AM or FM
modulation.

If you've added 1MHz to all the frequencies, and clipped the signal in
that form, the results are quite different than clipping the detected
audio. For a single input frequency, all the distortion products end up
above 2MHz and thus get removed by the filtering.

The distortion ends up being heavy on the IM distortion effects and light
on harmonics. For some reason, this seems to be easier on the hear.

I would say it is likely the same on IM. (Or at least there is no special
reason for it to be different.) The harmonics, as you say, are "gone." So the
total distortion is less.

In effect it works because it increases the S/N ratio at the RX'er by decreasing
the crest factor in a peak power limited TX'er. That is, the *average* TX power
is able to be driven higher.

 

[Ken Smith]

I read in sci.electronics.design that Ken Smith
about 'Instantaneous (analogue) compression of speech signals', on Fri,
7 Jan 2005:

Not in my experience, but that is with systems having wider bandwidth.
One would expect IM to sound worse, because the IM products are not even
vaguely harmonically-related to the fundamentals.

Yes, I've only ever seen this in voice grade communications circuits.

Perhaps added harmonics makes "s" harder to tell from "sh" and etc but the
IM stuff doesn't.

 

[gwhite]

What do you doubt? I've posted as much detail about the application -
induction-loop systems for use with hearing aids - as I can without
compromising certain interests.

The reason I doubt it is because every source I study says that it isn't so, and
it has been studied for hard and long. Clipping doesn't seem to hurt. It is a
secondary effect as best I can tell from all my readings. So as a secondary
effect I wonder how much it could improve things even if a special combo was
found to have some positive effect. That is, given what I know any positive
effect could only be small in the scheme of things.

Don't get me wrong -- if you've found something special, I think that's great
and I don't expect you to disclose it.

Agreed, although those studies are on signals that can be more seriously
degraded (not by the clipping but by other system characteristics -
bandwidth and noise) than those I'm concerned with.

RF/IF clipping is not an option. The amplifiers concerned are analogue,
with transformer/rectifier power supplies. AS such they need minimal EMC
assessment and usually no testing. Introducing RF and/or digital
processing changes the situation greatly and involves significant extra
cost and development time.

I realize you ruled the method out as true RF/IF operation. There is the
baseband envelope clipping method which is absent RF/IF translation or digital.
I don't care what you do. Personally I think I kinda like the tanh method if
you can make it work like you want -- it is really simple.

 

[Jim Thompson]

[snip]

RF/IF clipping is not an option. The amplifiers concerned are analogue,
with transformer/rectifier power supplies. AS such they need minimal EMC
assessment and usually no testing. Introducing RF and/or digital
processing changes the situation greatly and involves significant extra
cost and development time.

I realize you ruled the method out as true RF/IF operation. There is the
baseband envelope clipping method which is absent RF/IF translation or digital.
I don't care what you do. Personally I think I kinda like the tanh method if
you can make it work like you want -- it is really simple.

Diff pair plus OpAmp, plus a DC loop to keep the diff pair balanced.

...Jim Thompson

 

[John Larkin]

For one channel of voice grade signal, I'd bet a PIC or 8051 based circuit
could do it. The tricky bit is the dynamic range of the ADC. It is easy
to get 24bits worth of analog dynamic range and harder to get that in an
ADC.

You're not likely to see much more dynamic range than 60 or so dB for
any real-world audio signal. So a 12-16 bit ADC should be good for
most apps. A DSP, or even a decent uP, could delay the data stream, do
an average or quasi-peak detection, envelope delay that some clever
smooth way, and multiply the delayed samples to compress the dynamic
range without bad artifacts. I'm sure it's being done already.

John

 

[Ken Smith]

It is actually done in the *transmitter* as a purposeful processing technique.

I've also seen it in the receiver as a way to prevent the headphones from
blowing your ears off. It was a simple clamping diode just in front of the
second detector stage. I suspect that if it is done in the transmitter,
and the receiver's clipping level is above the transmitters, it would
greatly reduce the distraction of noise spikes.

FM and AM don't per se "need" RF/IF clipping, but could use it effectively too.

Actually just about all FM receivers clip in the IF strip several times.
The ratio detector also effectively clipps the RF too.




------------>!----+--/\/\/----
( ! !
( --- !
+-- ---C1 +---- Audio
( ! ! !
( ! ! !
- ! ------!<-----+--/\/\/----
!
--------



C1 in my sketch is big enough that the voltage on it does not change at
audio frequencies.


[...]

I would say it is likely the same on IM. (Or at least there is no special
reason for it to be different.) The harmonics, as you say, are "gone." So the
total distortion is less.

Yes the total is lower, but I suspect tha in real circuitst the IM is in
fact slighty higher for the same amount of amplitude compression. When
you clip at baseband frequencies, you AM modulate one component with
another. At RF frequencies, there is sure to be some phase modulation
going on too.

 

[Ken Smith]

You're not likely to see much more dynamic range than 60 or so dB for
any real-world audio signal.

Yes, the quietist to loudest range is likely to be about 60dB or so, if
that is sense in which you mean dynamic range. But at the low end (when
someone wispers), we still need a few bits in the ADC. 16 bits would most
likely work if we oversampled and could count on noise to smear out the
artifacts at low amplitudes. It ends up being a trade off between bits
and speed.

 

[John Woodgate]

I read in sci.electronics.design that John Larkin <[email protected]>
wrote (in <[email protected]>) about
'Instantaneous (analogue) compression of speech signals', on Fri, 7 Jan
2005:

You're not likely to see much more dynamic range than 60 or so dB for
any real-world audio signal. So a 12-16 bit ADC should be good for most
apps. A DSP, or even a decent uP, could delay the data stream, do an
average or quasi-peak detection, envelope delay that some clever smooth
way, and multiply the delayed samples to compress the dynamic range
without bad artifacts. I'm sure it's being done already.

Yes, this sort of thing is kept proprietary and heavily protected by
patents.

 

[John Woodgate]

I read in sci.electronics.design that Jim Thompson
4ax.com>) about 'Instantaneous (analogue) compression of speech
signals', on Fri, 7 Jan 2005:

[snip]

RF/IF clipping is not an option. The amplifiers concerned are analogue,
with transformer/rectifier power supplies. AS such they need minimal EMC
assessment and usually no testing. Introducing RF and/or digital
processing changes the situation greatly and involves significant extra
cost and development time.

I realize you ruled the method out as true RF/IF operation. There is the
baseband envelope clipping method which is absent RF/IF translation or digital.
I don't care what you do. Personally I think I kinda like the tanh method if
you can make it work like you want -- it is really simple.

Diff pair plus OpAmp, plus a DC loop to keep the diff pair balanced.

Do you mean feeding the op-amp d.c output back through a potential
divider to the base of the tail transistor?

I've now investigated the tanh(sin(x)) function in Mathcad. For 3 dB
reduction in peak voltage, and the transfer function for the long-tailed
pair normalized to y = tanh(x), the peak input voltage has to be 1.15
normalized volts, and this gives about 9% third harmonic and 1% 5th. An
interesting result is that the harmonic spectrum is nearly linear on a
decibel axis, the nth harmonic level being nearly -10(n-1) dB. n is
always odd, of course, for symmetrical limiting. This linearity applies
for any input signal amplitude that I have tried, but the scale factor
changes, of course.

3 dB doesn't sound much, but it halves the amplifier power required.

Now to find what it sounds like on real signals. I have a breadboard all
ready!

Incidentally, I obtained those results by using traditional 'analogue'
Fourier analysis. I tried to use the Mathcad 'fft' function, using 32
samples, carefully arranged so that the input waveform started and
finished at zero amplitude. It gave me results which I did not believe,
including a succession of even harmonics of significant but suspiciously
similar amplitude and a fundamental component much larger than 1. I
thought that maybe 32 samples was too few, so I tried 1024. The results
were even worse; the fundamental component amplitude exploded to about
17! I'm sure there is an explanation, but it concerns me that it would
be very easy to get a wrong answer, using 'fft', that wasn't so
obviously wrong as to ring alarm bells.

 

[Jim Thompson]

I read in sci.electronics.design that Jim Thompson
4ax.com>) about 'Instantaneous (analogue) compression of speech
signals', on Fri, 7 Jan 2005:

John Woodgate wrote:
[snip]

RF/IF clipping is not an option. The amplifiers concerned are analogue,
with transformer/rectifier power supplies. AS such they need minimal EMC
assessment and usually no testing. Introducing RF and/or digital
processing changes the situation greatly and involves significant extra
cost and development time.

I realize you ruled the method out as true RF/IF operation. There is the
baseband envelope clipping method which is absent RF/IF translation or digital.
I don't care what you do. Personally I think I kinda like the tanh method if
you can make it work like you want -- it is really simple.

Diff pair plus OpAmp, plus a DC loop to keep the diff pair balanced.

Do you mean feeding the op-amp d.c output back through a potential
divider to the base of the tail transistor?

I don't know if you need a divider or not, but definitely a low-pass.

I've now investigated the tanh(sin(x)) function in Mathcad. For 3 dB
reduction in peak voltage, and the transfer function for the long-tailed
pair normalized to y = tanh(x), the peak input voltage has to be 1.15
normalized volts, and this gives about 9% third harmonic and 1% 5th. An
interesting result is that the harmonic spectrum is nearly linear on a
decibel axis, the nth harmonic level being nearly -10(n-1) dB. n is
always odd, of course, for symmetrical limiting. This linearity applies
for any input signal amplitude that I have tried, but the scale factor
changes, of course.

3 dB doesn't sound much, but it halves the amplifier power required.

Now to find what it sounds like on real signals. I have a breadboard all
ready!

Incidentally, I obtained those results by using traditional 'analogue'
Fourier analysis. I tried to use the Mathcad 'fft' function, using 32
samples, carefully arranged so that the input waveform started and
finished at zero amplitude. It gave me results which I did not believe,
including a succession of even harmonics of significant but suspiciously
similar amplitude and a fundamental component much larger than 1. I
thought that maybe 32 samples was too few, so I tried 1024. The results
were even worse; the fundamental component amplitude exploded to about
17! I'm sure there is an explanation, but it concerns me that it would
be very easy to get a wrong answer, using 'fft', that wasn't so
obviously wrong as to ring alarm bells.

I did it in PSpice and saw the same thing as you got with 'analogue'
Fourier analysis.

...Jim Thompson

 

[Jim Thompson]

I read in sci.electronics.design that Jim Thompson
4ax.com>) about 'Instantaneous (analogue) compression of speech
signals', on Fri, 7 Jan 2005: [snip]

Diff pair plus OpAmp, plus a DC loop to keep the diff pair balanced.

Do you mean feeding the op-amp d.c output back through a potential
divider to the base of the tail transistor?

I don't know if you need a divider or not, but definitely a low-pass.

[snip]

John, Circuit E-mailed to you, with an error in it... "quicky" is
asymmetric in the two "halves" gain

But you can fix that easily... I just mildly bungled the math, doing
it in my head :-(

...Jim Thompson

 

[gwhite]

[...]
I'm puzzled by "RF/IF" clipping. How does that work to improve the
demodulated audio?

In a single side band reciever:

It is actually done in the *transmitter* as a purposeful processing technique.

I've also seen it in the receiver as a way to prevent the headphones from
blowing your ears off.

Basically someone spent too much money on the headphone PA. More seriously,
sure, it is the "best known way to clip," but rarely applied in this manner.

It was a simple clamping diode just in front of the
second detector stage. I suspect that if it is done in the transmitter,
and the receiver's clipping level is above the transmitters, it would
greatly reduce the distraction of noise spikes.


Actually just about all FM receivers clip in the IF strip several times.
The ratio detector also effectively clipps the RF too.

------------>!----+--/\/\/----
( ! !
( --- !
+-- ---C1 +---- Audio
( ! ! !
( ! ! !
- ! ------!<-----+--/\/\/----
!

Obviously, and the RF of the TX'er often clips too. That's why I said it must
be "re-basebanded" (taken back to audio to FM/PM modulate, after processing) for
FM. FM is a "constant envelope system." Think about it. Clippers in FM are
pre-modulator for the purpose of limiting peak deviation and thus allowing
higher average deviation. No one is saying "AM a FM system."

[...]

I would say it is likely the same on IM. (Or at least there is no special
reason for it to be different.) The harmonics, as you say, are "gone." So the
total distortion is less.

Yes the total is lower, but I suspect tha in real circuitst the IM is in
fact slighty higher for the same amount of amplitude compression. When
you clip at baseband frequencies, you AM modulate one component with
another. At RF frequencies, there is sure to be some phase modulation
going on too.

I doubt that. You can "IF clip" at 20 kHz for voice grade systems, if you
want. (A common frequency is 455 kHz, which is practically DC.) The IM should
be the same. For example, term third order term (for two-tone) is:

a3*[cos(w1*t) + cos(w2*t)]^3

This is a general result--no bias regarding what the actual frequencies are.
Memory effects should be vanishing for the typical low frequencies of
processing.

 

[Fred Bartoli]

Does anyone here have any experience of instantaneous (analogue)
compression (aka soft clipping) of speech signals? I've been doing a
little work on it but I'm unable to judge the resulting sound quality.
Why do treble boost controls no longer have any audible effect for me?
(;-)
--

John,

I've not followed the whole thread so I don't know whether sb proposed this
or not.



10n
2.2K
___ || ___
.-|___|---||---+--|___|--+ .--------.
| || | | | |
| ___ | 15K | | |
-+--------|___|-+ | | |
| | | |\| | 15K
15K | |\| | 1K '--|-\ | ___
'---|-\ | ___ | >--'-|___|--+-----
| >--+-|___|--+--+----|+/ |
.---|+/ | | |/| |
| |/| | | ---
=== - V 10n ---
GND 2 diodes ^ - |
| | |
(or diode monuted BJTs) | | ===
====== GND
GNDGND

(created by AACircuit v1.28 beta 10/06/04 www.tech-chat.de)


I didn't tried it in real because I don't have what's required, but in
simulation it has some interesting effects.

I guess 1kHz is about a good corner frequency but of course you can adapt
it.

 

[John Woodgate]

I read in sci.electronics.design that Fred Bartoli <fred._canxxxel_this_
bartoli@RemoveThatAlso_free.fr_AndThisToo> wrote (in <41e25e76$0$6635$62
[email protected]>) about 'Instantaneous (analogue) compression of
speech signals', on Mon, 10 Jan 2005:

I've not followed the whole thread so I don't know whether sb proposed
this or not.

I've got something similar to that at present, although it has *bass-
cut* pre-processing rather than treble-boost. Jim Thompson's tanh
limiter does seem to have advantages, but I'll try your pre-processing
first, because that involves fewer changes to my breadboard.

I planned to do some work on it today, but of course, those damned
'clients' have intervened. -(.

 

[gwhite]

Agreed, although those studies are on signals that can be more seriously
degraded (not by the clipping but by other system characteristics -
bandwidth and noise) than those I'm concerned with.

Well that was my point when I said I doubted the application here. The
*traditional* case for it doesn't seem to exist; the transmission path doesn't
have those "serious degradations" according to you. Nor have you stated that
you have a peak limited system. Again, I realize you are claiming a new or
uncommonly known aspect.

So giving the "benefit of the doubt" to your IP, I wonder if you have considered
preemphasis before clipping and then deemphasis thereafter (unless a boost is
desired anyway). It is claimed that harmonic distortion from clipping can
"cover" the high frequency formants and reduce intelligibility, and in this way
preemphasis helps. (As a side note, this lack of harmonics covering high
frequency formants may help explain why RF/IF clipping is superior.)

 

[Ken Smith]

I doubt that. You can "IF clip" at 20 kHz for voice grade systems, if you
want. (A common frequency is 455 kHz, which is practically DC.) The IM should
be the same. For example, term third order term (for two-tone) is:[/QUOTE]

Please post a design for a clipper at 455KHz that will clip the IF strip
signal, without phase modulating. You have a budget of 1 diode, 1
resistor and 1 capacitor added to the design. You are not allowed to
lower the normal performance.

I think you will find it simply can't be done. At least one real SSB
reciever circuit had the parts suggested.

a3*[cos(w1*t) + cos(w2*t)]^3

This is a general result--no bias regarding what the actual frequencies are.
Memory effects should be vanishing for the typical low frequencies of
processing.

Not if said clipper is connected to a tuned circuit.