Multi Band Compressor

[Ian VK3KRI]

Hello All,
I'm currently building a home brew audio processor. One of the stages is a multi-band compressor. However I have come across an interesting issue, that those of you who have used real (ie commercial) devices with multiband compressors may be able to help me with.

The basic idea is that I have 6 bandpass filters strategicly spaced to cover about 100hz to 10khz. The crossover points between each filter are at the -3dB points on the filters edges. Thus at the crossover point we have a signal -3dB down through the low filter and another signal -3dB down through the high filter. Combining these we get a signal 0dB down and hopefully a flat response (yellow line) across the audio range as in the image below...

However - when the outpus of the bandpass filters are shoved into a compressor for each band , unexpected (but explainable) things happen. The compressor 'flattens out' the response of the filter. so that, although the output of the filter may be 3dB down at the crossiver point from the centre of the filter, the comressor dutifully reduces the dynamic range and makes the difference between the crossover point and the center of the filter less by an amout equal to the comression ratio. In the sweep below the comression ratio is almost infinity and there is a 3dB peak at each crossover point.

The question (eventually) is, do commercial units do this? If they do do it is it noticable? I suspect it isn't, because a slow audio sweep is not normal programme material . It doesn't seem too bad when I listen to it, but I may be fooling myself.

I have thought that a solution is to follow the compressor with another identical filter which will shape the output of the compressor, But this causes the response with no compression to be non-flat.

Anyway, any ideas are welcome.
                                           Ian VK3KRI

[AB2EZ]

Ian

Yes... I would expect to see exactly what you are seeing if the input sine wave level is above the threshold for compression to kick in. I noticed that my Orban processor does "interesting" things when I drive it with a sine wave. [The Orban processor has a built-in sine wave generator that bypasses the processing... but I'm referring here to injecting a sine wave from an external generator into the processing chain]. You will also see that with a sine wave input, the compressor is producing harmonics that are the consequence of the non-linear processing

With a complex non-linear processing chain, I think you are correct in using subjective testing with real program material to judge the performance and to adjust the settings.

Of note, Orban includes a complete schematic in the manual for its processor (i.e., they do not consider the hardware to be proprietary)... but they do not (of course) disclose the details of the processing algorithms... which are embodied in software.

Best regards
Stu

[AB2EZ]

Ian

As an example of what I referred to as "interesting" things...

After my last post, I injected a sine wave from an external oscillator into my Orban... and compared the input to the output as I varied the frequency from about 100 Hz to about 4 kHz. I didn't attempt to experiment with different settings on the Orban's processing... so both the automatic gain control (broadband) and the multiband compressor were active. I didn't see a significant change in the output of the Orban (the output of the signal generator is fairly constant as the frequency is changed) as I slowly changed the frequency...

But... I saw a dramatic change in the amount of third harmonic that was present at the output of the Orban. At frequencies below 700 Hz and above 2 kHz, there was negligible third harmonic visible (more the 45 dB down from the fundamental). Between 700 Hz and about 900 Hz the third harmonic at the output of the Orban (but not at the output of the audio generator) rose to about -20dB relative to the fundamental. It stayed flat at about -20dB down until about 1500 Hz, when it began to drop off again.

Thus, for the particular settings I was using (the settings I use with my RE-27), and for the particular input level I was using, it appears that between about 700 Hz and 2000 Hz one of the multiband compressors is active... but outside of that frequency range, the AGC is doing the work of keeping the output level constant.

Stu

[flintstone mop]

Interesting Thread. I guess shooting single sine waves into a processor will do strange things. How does your processor work with a voice or music input? I had an Inovonics MAP 231(?) and if you push the gain of one of the bands too high IT would "take control" of the final output process. All of those bands eventually come to central collection point for the audio. And the final touch-up of compression/limiting takes place there. Lottsa people like that extra bass in the AM signal but it's going to kill your chances of higher modulation of the overall audio signal out of the transmitter. I see the modulated waveform of WBCQ and the extra bass that is set-up in their audio chain kills the overall loudness of the station.The pipe line is not that big out of a transmitter AM or FM. There are technical limitations to each type of transmitter (A commercial FM transmitter seems to be limitless for amount of FM....I was playing around in a smoked up state at a college FM station, 1970's and the mod monitor was pegged at 150% and it just kept getting louder and louder.) and then there's the FCC.
Fred

[John K5PRO]

Another thing to watch for is how the phase at the filter corners (-3 dB) affects the response when they are summed back together. When I designed a multi-band compressor in 1992 (which was never built, only modeled and schematic done), I found that the summing point of the various bands wouldn't add up to a constant voltage if the circuit was swept, if one wasn't careful about what sort of LP, BP and HP elements are used. I made a 3 band unit. Using Microcap (a Spice program with nice schematic entry), I could play with filters until I was sick of them, and until I hit a specific topology, the result wasn't what was expected. I was sweeping them using the stimulus menu in the program. But the theory was exactly what was expected in practice. Crossover networks for loud speakers and bi or tri-amplified component amplification systems have the same concerns. Orban did it right in his boxes. Some companies didn't think about this, just using cookbook designs, with predictable results.

[Ian VK3KRI]

Another thing to watch for is how the phase at the filter corners (-3 dB) affects the response when they are summed back together. When I designed a multi-band compressor in 1992 (which was never built, only modeled and schematic done), I found that the summing point of the various bands wouldn't add up to a constant voltage if the circuit was swept, if one wasn't careful about what sort of LP, BP and HP elements are used. I made a 3 band unit.

Yes I tried higher order filters, but the phase alignment was rather screwy resulting in a very ragged summed result. The actual filters ar 12dB/Octave which are the same as in the analog Optmod 9000 which is where I have drawn my 'inspiration'.

This is actually a DSP processor - all done in software PC, which makes fiddling with parameters real easy.  Which is probably just as well, because at this stage I'm not really sure how much the multi band comressor really does for voice at this stage. Oh well, plenty more deveolpment time to come...

                                                                          Ian VK3KRI
 
                                                                 


 

[Ian VK3KRI]

Ian

As an example of what I referred to as "interesting" things...

After my last post, I injected a sine wave from an external oscillator into my Orban... and compared the input to the output as I varied the frequency from about 100 Hz to about 4 kHz. I didn't attempt to experiment with different settings on the Orban's processing... so both the automatic gain control (broadband) and the multiband compressor were active. I didn't see a significant change in the output of the Orban (the output of the signal generator is fairly constant as the frequency is changed) as I slowly changed the frequency...

But... I saw a dramatic change in the amount of third harmonic that was present at the output of the Orban. At frequencies below 700 Hz and above 2 kHz, there was negligible third harmonic visible (more the 45 dB down from the fundamental). Between 700 Hz and about 900 Hz the third harmonic at the output of the Orban (but not at the output of the audio generator) rose to about -20dB relative to the fundamental. It stayed flat at about -20dB down until about 1500 Hz, when it began to drop off again.

Thus, for the particular settings I was using (the settings I use with my RE-27), and for the particular input level I was using, it appears that between about 700 Hz and 2000 Hz one of the multiband compressors is active... but outside of that frequency range, the AGC is doing the work of keeping the output level constant.

Stu

Could that be the 'smart clipper' or whatever doing something? Not that I know anything about any of these processors apart from downloading the manuals. The ultimate test is , 'does it sound better' or 'is it more copyable' or both. I guess I'll be happy if I can do either, and I'll be real happy if I can do both!.
                                                                Ian VK3KRI

[AB2EZ]

Ian

Agreed on the subjective quality being the measure of success!

As you know, symmetrical compression of a sine wave will produce odd harmonics. That is why I suspect that the measurement results I described above are associated with the differing amounts of compression and AGC that are taking place at different frequencies (different bands of the 5-band compressor). Although Orban does not provide much detail about the alghorithms in the manual... the manual does include a discussion of how one can trade off between using more AGC and less compression or vice versa. From the manual of the Orban 9200: "Moderate AGC + light compression produces and open, natural quality with automatic re-equalization and increased consistency of frequency balance. Moderate AGC + heavy compression results in a 'wall of sound' effect that will maximize loudness and coverage but which may result in listener fatigue"

[Bacon, WA3WDR]

Compression after equalization will alter the apparent frequency response, tending to push down increases in output, and maybe pulling up decreases in output.  Testing something like this might involve pink noise input, and spectral analysis of the output.  Even then, it would need to be done over a wide range of amplitudes, because of the possible dynamics of such processing.

Possibly a sine wave at a low level could show the uncompressed frequency response, and a sine wave at a high level might show the limited output spectrum. The limited output spectrum might roll off at the top end, in order to work better with high frequency pre-emphasis in the transmitter.

-20 dB is a lot of third order distortion from modern audio gear.  It should be visible on an oscilloscope.

Peak limiting is not the same as peak clipping.  In general, peak limiting refers to fast compression to a fixed peak output level, but it is not necessary to clip in order to do this.  A steady sine wave should not be seriously distorted, but there is often transient clipping on sudden loud sounds while the AGC system is reducing audio gain.  The variable gain amplifiers usually add a small amount of distortion, but it is generally not the same as waveform clipping.  It is usually just general transfer curvature non-linearity, such as mild crossover distortion, etc.  It is possible to include some peak clipping by design, if desired.

[WBear2GCR]

Ian,

If you get to use DSP routines, perhaps you should try something besides the 12db/oct. One of the problems with second order filters is that the passband and the stop band are 180 out of phase...

Consider a 4th order Linkwitz-Riley out of the pantheon of "stock" filter alignments.
And, if the thing has enough power, a 'brick-wall' filter would be nice.
My limited understanding of DSP filters says that it is possible to make very high order filters without attendant phase shifts?

But I'm interested in what you arrive at - please keep us posted?

          _-_-WBear2GCR

[Bacon, WA3WDR]

Ian, what kind of filters are you implementing in the DSP?  Are they FIR or IIR?  I've worked with FFTs in DSP, but not audio equalizers.

Without compression, if the filters are shifting phase +/-45 degrees at the crossovers, and if they are 0.7071 (-3 dB) at the crossovers, then at the crossovers that would put them 90 degrees apart, and their vectors would combine to 1.0, and they would be pretty much flat.  But with the drive up, and the compression holding the output at 1.0 instead of 0.7071, with the same phase difference of 90 degrees, the vector sum at the crossover would be 1.414, which would be a 3 dB bump such as you are seeing.  And I think that if you had filters with no phase shift, rolling off to 0.5 at the crossover, then compression of the frequency bands would create even worse peaks of about 6 dB.

Probably the compression should be done differently.  Taking the +/- 45 degree phase, -3 dB crossover filter example, if instead of a gain control loop, the level for the gain control of each band was taken from separate filters having unity gain in the center, and having 3 dB peaks at the crossover frequencies and then falling off on either side, and the gain control for each band was then 'fed forward' to gain scaling following (or preceeding) the existing band filters, then as a sine wave or a narrow-spectrum sound went into the crossover region, gain would be reduced further in both bands passing it, and the recombination after compression would result in flat response.

The spectrum of voice and other typical sounds is usually fairly wide, and the ripple seen with extremely narrowband signals would probably not be a huge issue, but I think that the double-humped gain control filter response would be a better way to go.  And DSP opens so many possibilities that there are surely better ways yet, maybe variable auto-'notching' that contours itself to relatively high amplitude spectral concentrations.

Also (curiosity) - how are you doing the compression?  I think that you are reducing gain, at an adjustable compression ratio, when the level exceeds a threshold.  Are you looking at RMS, or peak, or something else?  I am guessing peak.  Also, what are you doing with the time constants?

[Ian VK3KRI]

Ian, what kind of filters are you implementing in the DSP?  Are they FIR or IIR?  I've worked with FFTs in DSP, but not audio equalizers.

The filters for the multiband part are IIR. And I should point out that I am using existing filter libraries, as I figure there are pleny of people out there who know more about coding them than I do! I just plug in the parameters and and they work.

Without compression, if the filters are shifting phase +/-45 degrees at the crossovers, and if they are 0.7071 (-3 dB) at the crossovers, then at the crossovers that would put them 90 degrees apart, and their vectors would combine to 1.0, and they would be pretty much flat.  But with the drive up, and the compression holding the output at 1.0 instead of 0.7071, with the same phase difference of 90 degrees, the vector sum at the crossover would be 1.414, which would be a 3 dB bump such as you are seeing.  And I think that if you had filters with no phase shift, rolling off to 0.5 at the crossover, then compression of the frequency bands would create even worse peaks of about 6 dB.

Thats sort of the conclusion I came to. Certainly the higher order filters combined rather oddly, because of the phase shift at the -3db point not being 90 degrees. The current filters combine nicly under 0 dB comression conditions, so I'm happy to stay with that for the moment.

Probably the compression should be done differently.  Taking the +/- 45 degree phase, -3 dB crossover filter example, if instead of a gain control loop, the level for the gain control of each band was taken from separate filters having unity gain in the center, and having 3 dB peaks at the crossover frequencies and then falling off on either side, and the gain control for each band was then 'fed forward' to gain scaling following (or preceeding) the existing band filters, then as a sine wave or a narrow-spectrum sound went into the crossover region, gain would be reduced further in both bands passing it, and the recombination after compression would result in flat response.

I like your thinking here. Feed forward filters to the gain control stage with an 'overcoupled' double hump response would certainly do the trick. I'll certainly put that on the to-do list. Just dpends how expensive they are to impement.

Also (curiosity) - how are you doing the compression?  I think that you are reducing gain, at an adjustable compression ratio, when the level exceeds a threshold.  Are you looking at RMS, or peak, or something else?  I am guessing peak.  Also, what are you doing with the time constants?

The front end ALC is effectively a VERY slow limiter - 200mS Attack, about 5 Seconds decay which seems to track input level changes quite well. However I will probaly change this over to detecting RMS rather than peak as it does wind back the gain a bit on music with repetitive peaks 
Its a simple threshold detector so its infinte compession, but with such a slow time constant it doesn't effect dynamics at all - hopefully. Also its gated so gain freezes at low input levels.
The compressors on the mutiband part are  some more bits grabbed out of a library and are also peak limiters but with a 5mS attack and 100mS release. These need to be a bit slower I think. I'l probaly stick something in there thats tunable as to the compression ratio snd compression threshold. The orban manuals say they are using diffents setings for the compressors on each band, so that could be some task to tune!

The final gain control stage is planned to be a look ahead peak limiter (not  clipper) with fast attack release times and assymetric thresholds. I'll be intersted to see if that actually works!

Also in there is/will be some high frequency pre-emphasis, variable high/low pass filters for switching to 'space-shuttle' audio.    Its an interesting learning experience if nothing else

                                            Ian VK3KRI
   

[Bacon, WA3WDR]

Very cool.  This is making me think about getting back into my old 56002 EVM, or maybe the 2182(?) EVM that I got, or doing it on the PC.

I agree about the 100 ms release time being too short.  All sorts of things go on with time constants, and if you can get a combination of several at once, it works better.  On wideband peak limiting compressors, a super fast one works like a peak clipper, because it responds to the waveform literally in real time.  A fast attack-release, overriding a slower attack and release with a few dB less gain reduction, works pretty well.  On an analog unit, this is done with a larger capacitor connected to the gain control bus through a series resistor.  The larger cap charges slowly, and then its slow effect on gain is reduced by the voltage division by the series resistor and the resistor to ground on the gain control bus.

I found that little 'tick' sounds don't affect the slow section, but they can hang the fast section, and a slower attack on the fast section compromises peak control.  This can really affect speech when strong equalization is used.  So I put a variable resistor in series with the fast section capacitor to slow the fast attack a little, and added a small capacitor across that resistor.  That added some super fast attack release, on top of fairly fast, on top of reduced-effect slow.  The super-fast section catches the clicks and clips them, and then releases in a few milliseconds.  The transient clicks are distorted, but that's not very noticeable, and they don't hang the fast section, and it sounds OK.  This is a compromise because it gets into transient distortion, but it controls peaks better, and it sure makes speech sound louder.  So I made the resistor in series with the fast section variable, and I can turn it up and down as needed.

Just some thoughts.

[W1DAN]

Ian:

Very cool what you are doing!!!

I agree with Bacon's ideas. When you are below any AGC/limiting threshold, the frequency response of the filters added should be flat. When audio is driven to where the gains are changed you will have peaks and valleys at the output. This is why a limiter is added to the output. If everything added in phase and amplitude, you would not need a final limiter for transmitter modulation protection.

I have read and studied many of Orban's processors manuals and patents and have learned some of his tricks.

Orban uses clipping much more than most other processor manufacturers. He gets way with it as he is able to reduce the resulting harmonic distortion after clipping. I think that you can get only so much loudness without clipping. Consider a careful and light clipper and brick wall filter after your final limiter.

I have built an analog 3 band compressor and limiter in the 1990's and have read and played with DSP. I have not built a DSP limiter yet, so I am very interested in how you are doing this (i.e what platform...PC or DSP chip).

Thanks for posting the question on this board. Keep us posted!!

Fun stuff!

73,
Dan
W1DAN
 

[KA1ZGC]

Ian,

Maybe this has already been touched on (I gave the thread a cursory read), but one thing you'll want to look into is your first stage of gain control.

This doesn't address your passband ripple per se, but can avoid other problems in the future.

Since you're trying to replicate the Orban functionality in DSP, I assume you are doing a broad-band (i.e. not-frequency-specific) compression at your first stage. The question then becomes not so much one of attack time than as your triggering signal.

You will probably want to sidechain your broadband gain control such that it's reacting more to bass and low-mids than anything else. Without this, you will find your bass and low-mids being compressed by percussive sibilant sounds, such as your Ps and Ts. You want to let those percussive sibilant products pass unabated through the broadband compression, and deal with them in the appropriate band-specific portion when they get there.

This way, you avoid pumping and breathing resulting from normal speech products, such as you typically get from more straightforward compressors.

What this amounts to in software is placing the appropriate low-pass filter routine between the system audio input and the control input for the broadband compression (not the compressor audio path, but the compressor control path).

Sounds like a great project, Ian. It's one of those things I always wanted to do, just never found the time. Good luck with it!

--Thom
Killer Audio One Zero Grit Compressor

[steve_qix]

I use a home-brew 5 band processor.  I designed and built this back in 1979, so my memory is slightly foggy on the exact details, but I *do* remember eliminating that crossover hump by slightly moving the filters' 3 dB down points, so I got a pretty much flat overlap.

I'm still using this processor today, almost 30 years later.

I use a fast attack, slower decay.  The decay on the low bands is a bit higher than the decay on the high bands.

I use a JFET in the feedback loop of an op-amp for the control element for each band, and a full wave rectifier for each.

There are better gain cells available today than were available back 30 years ago, although the FET in the feedback loop works rather well .

Regards,

Steve

[Ian VK3KRI]

--Thom
Killer Audio One Zero Grit Compressor

An interesting set of phonetics, as one of the current areas of work is reducing 'noise' on the 'agc' line in the compressors/limiters.  I'd never given it much thought, but the 'voltage contralled amplifier' stage is baically a modulator, multiplying the input audio signal by the control voltage. While the control voltage is DC this is no problem, but one there is ripple (sylabalic or low audio rate)  or impulse (from fast attack) noise on the AGC line it will modulate the input signal . I am guessing that the products are the cause of the 'grit' on fast attack comressors.   Listening to the AGC signal I can certaing hear wideband noise from fast attacks, which I'v reduce by reducing attack time and doing dome primative low pass filtering on the AGC signal.

Ive been doing some local testing from TX into receiver with attenuators to get the RF signal down into the noise. So far the multi band compressor doesn't really seem to do much at that S/N ratio. However 15 dB of look-ahead peak limiting certainly adds some punch! IT does sound noticably comressed though.
                                                                        Ian VK3KRI
                     

[KA1ZGC]

The Orban units have a certain "grit" to them anyway, but that's more a product of the digital delay line(s) used to make the compression forward-acting.

The essence of any of these units is to have a relatively slow-attacking (like 200ms) broadband AGC as your first stage (again, with the controlling DC derived from energy below the spectral gap), then fanned out to the parallel filtered loops which have a slightly faster attack (66-75% of the broad attack time), then fanned back in to the final limiter (which, of course, has the fastest attack time of all). The output of the limiter is fed through a brick-wall low-pass filter to remove any clipping products the limiter may have introduced.

Careful tweaking of the first stage will afford you sufficient headroom in the following stages to avoid the effects of your passband ripple, and should also help your grunge factor. As long as your stages progress from slow attack to fast attack, the system should work like most multi-band compressors.

Another thing to consider is using five bands instead of six. With a broader sampling area for each section, you may find your control values find a more consistent vector sum, leading to less grunge. This would also help with the passband ripple issue, as you'd be widening your 3db points to compensate for the reduced number of bands.

Of course, I'm not entirely sure *where* you're hearing the wideband noise, or exatcly how you have the chain sequenced, but that's essentially what goes on in the hardware units.

How is the chain sequenced now? We've been talking about the multiband portion of the system; what is being done to the audio before and after the multiband processing? Therein may lie the key.

--Thom
Kilowatt Amplifier One Zero Grid Current

[Bacon, WA3WDR]

I found that asymmetrical peak limiting sounds much more natural than symmetrical peak limiting, when the compressor asymmetry is matched with the audio asymmetry.  I think this is because it allows asymmetrical waves to be higher in peak amplitude than symmetrical ones, and this allows them to be about equal in RMS power, and that is more like the normal sound.  Often the asymmetrical sounds are the emphatic ones, and if they are lower in RMS amplitude than vowel sounds, it causes the speech to have a sucked-out quality, especially in noise.

In baseball game broadcasts, symmetrical compression always seemed to punch a hole in the crowd noise when the announcer spoke, but the announcer's voice wasn't as loud as the crowd, and the effect was really irritating when the crowd got excited.  The fact that it was fast peak limiting only made the effect worse, by filling the time between syllables with louder crowd noise.  I once ran some of that kind of ball game audio through a fast asymmetrical peak limiter, and it sounded a lot better.  Probably 'phase randomization' would help that situation too, but the output is not as asymmetrical for amateur AM.

[Ian VK3KRI]

Of course, I'm not entirely sure *where* you're hearing the wideband noise, or exatcly how you have the chain sequenced, but that's essentially what goes on in the hardware units.

How is the chain sequenced now? We've been talking about the multiband portion of the system; what is being done to the audio before and after the multiband processing? Therein may lie the key.

--Thom
Kilowatt Amplifier One Zero Grid Current

I'm not hearing wideband noise on the audio, well not that I can notice it directly. Its noise on the AGC 'line' which I'm assuming will mix with the audio signal and cause some degradation of the audio. 

Currently the input AGC has an attack of 1/2 a second . Which reminds me , is there an industry standard way of defining attack/delay times. I always assumed a 100mS attack was what you got with a R-C combination with a 100mS time constant.  But squinting at schematics it appears that this isn't the case, or else my calculotor is wrong. Also it doesn't apply to compressors/limiters with use constant current sources/sinks to charge /discharge caps. with gives a linear voltage change with time.

Anyway currently the line up is:

- Differential input using L & R channels to try to reduce the $%$$% noise from the onboard soundcard.
- Variable high/low pass filters to band limit the audio appropriately for conditions
- Slow, Gated AGC. currently 0.5 sec attack 10 sec decay. (still using broadband level detector)
- some Pre-emphisis and EQ. This is where it is in an optimod 9000, but I'm not sure if this is the right place. as the multiband comressor 're-equalises' signal anyway. I don't need really needd  the EQ , but I do need pre-emph somewhere.
- 6 parallel band pass filters. Note that the top an bottom bands aren't high and low pass filters, all 6 are the same type of band pass. Im going to move all these down in frequency a bit as the bass end rolloff is too high at the moment.
- into 6  limiters.  These currently have quite fast attack - full gain to 0 gain in 50mS . The release times are  variable , but as the freqency band increases the release time is reduced in ratio to the frequency increase.  The bottom band is from 0db/S release to 20dB/sec release.
- these are summed back together  and shoved into a lookahead limiter which has a 5 ms time delay buffer and an attack time of 5ms so the signal can never exceed the limit (in theory) release is also 5mS. Tonights task is to convert this to an assymetric limiter.

The main issues at the moment are the noise floor of the sound 'card'  particarly with 20dB or so of compression. The one on the motherboard is pretty bad.
The other is delay through the whole system.  In theory this should only be about 10-20mS, but because  system occasionally decides it has more important things to do like flushing stuff to disk and other housekeeping stuff, I need to buffer a couple of hundered milliseconds audio to avoid underruns to the audio device.  This rules out monitoring  yourself off air ....
                                                                                     Ian VK3KRI




 

[W1DAN]

Hi Ian:

Good stuff!

You are doing very well on your project and am jealous!

The noise floor on sound chips on a mother board are not very good as you have discovered. Consider using an external USB audio interface or a $100-200.00 sound card. If you do alot of compression, the fact that you are running 16 bits may give you a little grunge on the very low level audio coming into your sound chip. Wonder of you can get 18-24 bits into your program with the proper sound card?

You can reduce the noise on the AGC "line" by low pass filtering it as you have done. Consider delaying the audio to be controlled by the same amount of delay the AGC filter has.

Are you making your gain changes at the zero crossing of the waveform?

Your pre-emphasis location is a good spot to have it.

Keep in mind the "orban" analog sound is due to the bucket brigade delay line used in the 9000, but more importantly Bob Orban uses a clipper at the output of each multiband compressor as well as at the point where the bands are re-combined. He also performs distortion reduction, generally below 2.2khz. This woudl be hard to reproduce digitally.

Digitally he does a main distortion cancelling clipper (and yes AGC is a modulator or mixer).

Hope this helps.

I enjoy your progress and wanna make a monoband compressor/clipper/LPF using an Analog Devices DSP Demo card someday.

73,
Dan
W1DAN

[KA1ZGC]

Ian,

If I were you, I would disregard the issues of soundcard deficiencies for now and focus on the software side of things. Once you've got your processor operating the best it can with the hardware you have, it's a fairly straighforward matter to transfer the software to a demo board (like the Analog Devices unit Dan mentioned, or perhaps Texas Instruments, there are several big names in this space).

Once you've got your processor on dedicated hardware, you get around the piss-poor DACs and clock lines bleeding into the audio, and you can then tackle any discrepancies that the soundcard may have masked. One or two reflashes may be all you'd need at that point.

Of course, the issue then becomes one of control, but once you've got the hang of burning the DSP chip, something like a Basic Stamp would be a simple and effective way of providing a user interface. There are lots of examples of Basic Stamps in exactly that kind of service publicly available on the net.

It sounds like you're almost there now, and you're heading down the right path to tackle your remaining issues. Keep us posted on your progress, and best of luck to you!

--Thom
Killer Album One Zappa's Greatest Compositions