Instantaneous (analogue) compression of speech signals

[John Woodgate]

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.

I'm not sure what you mean by 'peak-limited system' in this context,
since what I am asking about is peak limiting by another name, but there
is a considerable financial incentive to reduce the current and
compliance voltage requirements of the amplifier. I use those terms
because the induction loop load is reactive and is normally driven by a
current-source amplifier.

Again, I realize you
are claiming a new or uncommonly known aspect.

It is generally unknown: several people working in the field know about
it, but not in quantified terms. The IP is in the quantification.

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.

Yes, spectral conditioning is a known technique since the 'infinite
clipping' work of Licklider et al long ago. There are two essentially
different ways of doing it, and I have tried one. The other way has been
proposed in this thread, and I intend to try that as well.

 

[gwhite]

FT the signal

Raise each amplitude to the 5/7th power but don't change the phase

That's a general problem where I can't see how you've provided a method. How do
you propose a system resolve two tones and raise power individually. If this
were possible, we really could have amplifiers with *no* IM products. It has
never to my knowledge been accomplished and really seems a contradiction of
terms: non-linearity that has no non-linearity products. ???

The input to the system is random. "Tone frequencies" are unknown beforehand.
More specifically, the inputs are not even tones, and no reasonable filtering
method could have the needed resolution. The actual implementation of a
(·)^(m/n) *frequency domain* "clipper" needs a bit more discussion too.

iFT the new spectrum.

Okay, lets define a standard two-tone equal level signal (let's use analytic
signals for ease):

x(t) = (e^(j·w1·t) + e^(j·w2·t))/(2*pi)

These tones can be arbitrarily "close" or "far" apart.

FT'ing this:

F{x(t)} = X(jw) = dirDel(w-w1) + dirDel(w-w2)

where

dirDel(·) := the dirac delta function

Because applying the rational power function to each tone individually seems to
have no obvious general solution, we apply it to the input generally:

Y(jw) = (dirDel(w-w1) + dirDel(w-w2))^(m/n)

where m/n is some rational fraction; 5/7 if you like.

IFT'ing:

1 /inf
y(t)=invF{Y(jw)}= ----| e^(-j·w·t0)·(dirDel(w-w1) + dirDel(w-w2))^(m/n)dw
2·pi/-inf

I don't care if you evaluate in the frequency or time domain. It remains to be
shown how this will not produce distortion products, either IM or harmonic
(harmonic is simply a subset of IM anyway.) I would actually like to see the
transform of the integral anyway.

No new frequencies are created and no interaction between the amplitudes
has happened. This method has neither harmonic nor IM distortion.

But you wrote:

"The intermodulation distortion will not be made zero by this method."

"The intermodulation distortion will not be made zero by this method. If
the input has more than one frequency component, the distortion will be
much higher."

So unless I misunderstand you, there is a contradiction.

 

[Ken Smith]

Ken Smith wrote: [...]

FT the signal

Raise each amplitude to the 5/7th power but don't change the phase

That's a general problem where I can't see how you've provided a method. How do
you propose a system resolve two tones and raise power individually.

You run the signal into a DSP that does what I suggested. This forces you
to use a DFT but for material of limited length, the DFT is good enough.
The system has a very large input to output delay but it does not have to
produce IM distortion. This is not something that can be part of a
realtime system since the input to output delay is greater than the
material length. It is, however, something that could be done to the
material on a CD and a new CD written to be played at a later time.


[...]

But you wrote:

"The intermodulation distortion will not be made zero by this method."

That sentence is from a different context about a completely different
subject.

So unless I misunderstand you, there is a contradiction.

Yes, you've mix up two ideas that are not related.

 

[gwhite]

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:

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.[/QUOTE]


Okay. So buffer it. I've never seen an application of a single diode. Yes,
AM-PM would likely occur if a single diode (or even single-ended collector
clipper) was used. I really don't know what you have in mind with a single
diode.

 

[john jardine]

"Fred Bartoli"

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.

Wierd!.
Never had chance to play with a compressor, so ran a 2 second 8kb .WAV
through the above (an explosion) and listened to the result. What surprised
me was the output waveform had been compressed, yet every nuance of the
original waveform followed through to the output.
Ran the .WAV through 2 of the above circuits in series (X2 gain in
between). Even more compression even more of a 'levelled' explosion.
Think I'll stick 8 in series and listen to what happens. Maybe salt in some
bass cut varieties!.
regards
john

 

[Fred Bartoli]

"Fred Bartoli"

Wierd!.
Never had chance to play with a compressor, so ran a 2 second 8kb .WAV
through the above (an explosion) and listened to the result. What surprised
me was the output waveform had been compressed, yet every nuance of the
original waveform followed through to the output.
Ran the .WAV through 2 of the above circuits in series (X2 gain in
between). Even more compression even more of a 'levelled' explosion.
Think I'll stick 8 in series and listen to what happens. Maybe salt in some
bass cut varieties!.
regards
john

I didn't tried this but the sims let me expect something like this.
Can you run some other program through it, like speech and music ?

Maybe you have the opportunity to make a wav and post it back ?
I'd be curious.

For the simulation, run a 500Hz + 4kHz sin waves at various levels, and
observe the output, vs a simple R+diode clipper.
This one is amazing and can handle really **huge** overload levels. (I
didn't tried more than 2 sinus)

 

[john jardine]

"Fred Bartoli"

I didn't tried this but the sims let me expect something like this.
Can you run some other program through it, like speech and music ?

Maybe you have the opportunity to make a wav and post it back ?
I'd be curious.

For the simulation, run a 500Hz + 4kHz sin waves at various levels, and
observe the output, vs a simple R+diode clipper.
This one is amazing and can handle really **huge** overload levels. (I
didn't tried more than 2 sinus)

I've only a few *.WAVs on the PC, so I'll see if I can pull something from
the net.
I'll post the (single unit) before and after WAVs to ABSE.
(Input WAV is 8bit at 8k bits per second. Output WAV saved as 8 bits at
11kbits per second. Single channel only.
(Ps, Done via an LTSpice sim, so the diodes suffer perfectly symmetry)
regards
john

 

[Ken Smith]

Ken Smith wrote: [...]
clipper) was used. I really don't know what you have in mind with a single
diode.

A picture's worth 999 words:
This is the 1st example of this sort of circuit I saw.

B+
!
\
/
\ R1
/
!
----+-----+-----
!C1 ! ! -----+-----
--- ! ) ( ! To
--- ! ) ( --- Next stage
! --- ---) ( --- (detector section)
GND --- ! ) ( !
! ! ! -----+------
! ! !
! ! !
B+--!<--+-----------
D1 !
!
Plate


D1 clamps the positive swing of the tank circuit and thereby the amplitude
of the signal on it. Its from a GE communications radio, I had, that got
lost in the great flood.

R1 and C1 may have been already needed in the circuit for other reasons.
The circuit worked fairly well. The crashes of lightning etc were not too
much louder than the folking being listened to.

 

[Fred Bartoli]

I've only a few *.WAVs on the PC, so I'll see if I can pull something from
the net.
I'll post the (single unit) before and after WAVs to ABSE.
(Input WAV is 8bit at 8k bits per second. Output WAV saved as 8 bits at
11kbits per second. Single channel only.
(Ps, Done via an LTSpice sim, so the diodes suffer perfectly symmetry)

Ah, yes.
Thanks John, I forgot about that LTSpice capability.

 

[gwhite]

I'm not sure what you mean by 'peak-limited system' in this context,
since what I am asking about is peak limiting by another name, but there
is a considerable financial incentive to reduce the current and
compliance voltage requirements of the amplifier.

One of the "checkboxes" in radio work for justifying clippers is that of having
a "peak-limited system." I can explain what it means in that context, and then
I think one can transfer that idea wherever else appropriate.

The crest factor of speech signals is matter of concern for voice communication
systems. Speech is not deterministic and there are various statistical
standards for describing the crest factor. For this example, I'll simply use
the old Motorola 2-way radio standard of 13 dB for a 3-second average. (This is
actually a bit *lower* than most other standards I've seen.)

If such a signal is to be transmitted with fidelity, then the transmitter's
average power will be 13 dB below peak (instantaneous) power. Many/Most
transmitters are capable of delivering much higher average power, often PEP or
nearly so. So where this is true, we see the system constraint: for fidelity
reasons the transmitter is peak-limited, not average limited. It can provide
higher average power, but only at the cost of limiting peak amplitude (losing
absolute fidelity). If we provide 6 dB of clipping to the signal, then our link
budget (S/N) has straight-out improved by 6 dB because "non-severe" peak
clipping does not seem to harm intelligibility (but does affect subjective
quality). Clipping is an easy checkbox for this type of system.

Clippers are used in FM too. The fact that the FM TX'er typically runs at
constant PEP, may cause one to wonder why clippers are used there. In FM,
speech clippers simply limit peak deviation and have no effect on actual
transmit power. However, due to regulatory constraints, and the associated
RX'ers filters, the FM system is certainly peak-deviation limited, and that is
what it means there. It does increase end-to-end S/N here because the *average*
deviation can be made higher than it would be otherwise. In this way, 2-way FM
systems are also peak-limited.


In a sense, of course, all systems are peak limited. Traditionally it has
simply been a trade-off between fidelity under low noise conditions and
punch-through in high noise conditions.