Reverb Subculture

[seancostello]

Originally Posted by Froombosch

Thank you and Casey for your responces . Food for thought.

I should also add that plenty of "classic" reverb algorithms used linear interpolation for modulated delays.

[Casey]

Originally Posted by seancostello

No problem - just make sure that you adjust your delay lengths to take into account the block delay in the feedback path. For example, if you want a 1000 sample comb filter, and your block size is 32 samples, you would use a delay of (1000-32)=968 samples.

How you handle internal block delays is your own problem.

It is the alignment issue that must be attended to. Delays cannot be multiples of some block size.

So, and Dale alludes to this, some head and tail processing must take place to allow delays to be of arbitrary length.

Really, this is just for the record, I know you understand this!



-Casey

[Froombosch]

Originally Posted by seancostello

I should also add that plenty of "classic" reverb algorithms used linear interpolation for modulated delays.

Yeah i found that out also. ...

Harrie

[chrisj]

Originally Posted by Casey

I would have to look back, but I seem to recall that Rob King donated that sample.

Really? Well, you would be the one to ask- when these come out, may I put up mp3 clips using those dry samples (in other words, what's on this thread?) on Gearslutz? I think it would help people if they could hear a clip right in their browser, and I don't do that as often as I should. I'll give credit for the help this thread has been

[dale116dot7]

Originally Posted by Casey

How you handle internal block delays is your own problem.

It is the alignment issue that must be attended to. Delays cannot be multiples of some block size.

So, and Dale alludes to this, some head and tail processing must take place to allow delays to be of arbitrary length.

Really, this is just for the record, I know you understand this!

Processing by-block is probably fine, but actually, by-sample probably isn't that much more once the samples are moved into internal RAM - it should hopefully be whatever the programmer wanted to do, not necessarily a limitation of the hardware. Even processing by blocks can be tricky with SDRAM as the SDRAM page length probably doesn't line up with the desired delay line lengths.

What I would be tempted to do is make an internal RAM delay line equal to the desired delay line length (and making the basic delay line non-tappable except at the output end), then you subtract out however many 'blocks' you need out of the middle, leaving maybe 100 or 200 samples in the internal SRAM. Then you can DMA into and out of the middle of the array, although care must be taken about delay line wraparound, since the DMA controller must be programmed to reload its address register once it gets to the end. The modulus DSP addressing mode takes care of the code of an arbitrary length circular buffer, but the DMA controller may not be that sophisticated. If the delay is less than, say, 300 samples or so, then don't bother moving the data around - just leave it in SRAM. There would be a bit of work in the memory management control software, but it only needs to run to completion at least every 'N' samples where 'N' is the transfer block size (which may or may not be equal to the algorithm block size). But in this way, a delay line can be interpolated at the end (in SRAM), and it can be an arbitrary length. The algorithm that takes the reverb descriptor tables and allocates memory based on programmed reverb size would need to manage setting all of this up. You also could program sample-by-sample if you wanted to, or in blocks. In this way, the SRAM delay line acts as the arbitrary length adjustment, and also as a (manually maintained) data cache.

I know there are more ways to do this than this, but it should work ok. Another option is to have two circular buffers. The write buffer is, say, double the transfer block size. Once the write pointer gets halfway, write the first half. Once it wraps around, write the other half. The read buffer would be similar - the same length. The phase offset between the read and write pointers, plus the length of the bulk SDRAM delay would determine the delay line length. I can only think of maybe five ways of managing the SDRAM to SRAM data shuffle, and a couple of ways of doing the calculations in-place.

There are three main time delays in SDRAM that are important from what I can tell out of the datasheets. The first delay and the last delay are obvious - the open page and close page operations. The other delay is address supply to data available, and that applies to in-page accesses. And to make matters worse, if your block size aligns to a page length in SDRAM (which it should), that would be fine, but also for efficient access to SDRAM by the software, you either will have to misalign the writes (open and close two pages for one operation) or misalign the reads, unless the delay size is also a multiple of the block size (which also aligns with SDRAM page size). SDRAM also does not necessarily do well when you try to interleave writes and reads to the same bank, though that depends on the SDRAM controller built into the DSP.

This is interesting, I think one thing that's missing from people developing software these days is how the software actually has to mess with chips.

[Casey]

Originally Posted by dale116dot7

for efficient access to SDRAM by the software, you either will have to misalign the writes (open and close two pages for one operation) or misalign the reads.

And what about those pesky allpass filters that like to read AND write at both ends.



-Casey

[seancostello]

Could an FPGA be used to aid with getting data out of the SDRAM and into the DSP / local SRAM system? Maybe with modulo math?

Of course, if you have an FPGA doing the delay line management with the cheaper chips, it would be tempting to create a full reverb solution based around said FPGA.

This conversation is reminding me of having to deal with the 2126x SHARCs during the ADI days, before the 21369 came out. The 2126x chips had the memory access pins removed, due to some large customer requesting this. So all external memory access had to go through DMA, which was horribly slow. The block based processing helped a little, but not much. The Blackfin DSPs were coming out at that time, and were viewed as underachievers for audio by most of the higher ups at ADI. In our little group, we were able to get much higher performance out of the Blackfins than the 2126x for delay based algorithms. The 2136x series restored the memory access pins, thus saving face for the SHARC line.

[dale116dot7]

Originally Posted by Casey

And what about those pesky allpass filters that like to read AND write at both ends.

If your allpass is in-line (ie. one big memory loop and not several segments that are kept separate), then you should not need to read and write both ends simultaneously. If you consider the input to an allpass as coming from somewhere else (say, the end of another delay line that is totally separate), and consider the output as going to somewhere else (the start of another delay line that is totally separate), then you need to read feedback from the end of the allpass, write to the start of the allpass, and write the output to the head of its own delay line. You will waste some SDRAM this way, but SDRAM's pretty cheap and there's lots of it. Provided that there is at least one transfer block size of empty space in the SDRAM between delays, then there should not be an issue. Or am I missing something?

[cleartrueblue]

Originally Posted by Casey

I would have to look back, but I seem to recall that Rob King donated that sample.



-Casey

Could be wrong but I think that orchestral example is of a VSL mock-up by Berlin composer Andreas Koslik - posted on a German site when the M7 first came out. Actually, those files posted then are what sealed my purchase!

[chrisj]

What about the guitar clip? That's awfully well suited to a smaller room ambience.

[seancostello]

Just found this:

https://ccrma.stanford.edu/~dattorro/Griesinger.jpg

For those of us that were always fascinated by the Dattorro papers - and the story behind them - this letter sheds some light on the subject.

Interesting to read what Griesinger says about the Gardner reverbs as well.

[Froombosch]

Very interesting. Thanks for the paper

[seancostello]

Something I just noticed in Griesinger's letter:

"I know it is possible to decode algorithms, and have done it myself."

I wonder if this might be a reference to the Constant Density Plates in the 224 and 224X/L, with regards to the EMT250. Otherwise, I wouldn't know what Griesinger would have wanted to reverse engineer during his tenure at Lexicon.

Or maybe Griesinger is talking about decoding the older Lexicon algorithms, after the development machines were, um, lost.

[Froombosch]

Reverse engineering is a way to get some information on the famous old algos. Doing without is like reinventing and that is not the easy route. The shared information like in this thread is but also in the datorro papers are very valuable, because they give real insight.

I did start effectprocessor.com in 2004 to share some of the information I gathered.

[acreil]

I'm really happy to see an open discussion of this. I've read through the thread (and related threads) several times and carefully taken notes, and I'm trying to digest all the papers and patents I can find on the subject. I was going to just work on it by myself, but it seems more in the spirit of things to share my observations and questions.

I'm building reverberators and other effects in Pure Data. After a few initial experiments, a couple "cloned" algorithms, and some increasingly stupid ideas, I think I can safely clear all effect processors, hardware and software, from my wanted list. That probably sounds excessive, but my tastes tend toward the exotic, so I don't mind making things that turn single piano notes into underwater reverse bowed Balinese gong tone clusters. Making something that resembles an actual acoustic space seems quite difficult, but "outlandish" can be a lot more forgiving. I've only been messing with these for a few weeks but I'm really surprised at how quickly I'm able to make and refine them. As I become accustomed to the freedom I'm allowed here, I suspect even the most powerful Eventide/TC boxes would feel limiting.

Some ideas and things that I've tried:

1. Frequency shifters in feedback loops. I suspect Isao Tomita may have done this with a Bode frequency shifter and AKG BX20, results are interesting but it's perhaps too tempting to take it to total excess.

2. Dispersion filters- I saw that Sean participated in a paper modeling spring reverb (I don't have the full paper), and I saw a patent (4955057) suggesting 20-ish allpass analog phase shifter stages in a reverb feedback loop, which I tried. It's interesting in comparison to normal APFs, since an impulse is spread into a chirp, but it's sort of computationally intensive. Gary Kendall wrote a paper (The Decorrelation of Audio Signals and Its Impact on Spacial Imagery, 1995)
suggesting time variant decorrelating filters made from long chains of allpass biquads- maybe this is another option? I imagine sticking these inside normal APFs could be useful, computing power permitting. Anyone tried it?

3. That random delay Yamaha patent that Sean mentioned (553150)- I didn't expect a lot but I tried it with 16 delays and got interesting results. A single "comb" made this way can't handle much feedback due to the amplitude modulation of the crossfade, but modulated APFs and additional feedback paths (sort of a naive FDN) can actually make things quite smooth. Without feedback it can be used as a weirdly modulating "early reflection" FIR. It's really interesting and I'll have to play with it a lot more.

4. Long APF chain with random taps: I chained 16 APFs per channel and randomly crossfade between taps at each stage for output, feedback, etc. This can be used in a relatively sane way, or to make reverbs with a sort of dynamic "envelope", building slowly or quickly or panning from right to left or whatever. It's pretty cool so far, but I need to extend it.


Some questions and things:

1. I've had trouble finding some papers. I have a healthy pile of user manuals, service manuals, patents, papers that the authors have made available on their websites, DAFX ones, etc. but I'm missing some Computer Music Journal/JAES/etc. ones that are probably fairly important- in particular Schroeder, Blesser, Jot, Gerzon, and the Agnello one describing the Eventide SP2016. Anyone got these handy?

2. I've gotten the impression that it's best to modulate "in the loop" APF delay times (one paper suggested modulating feedback coefficients, though Sean described poor results), but are there limits to this? Is it a logical conclusion, given adequate processing power, to just "modulate everything"? Does it make more sense to use one n-phase sinusoidal or triangular modulator for everything, or to break it down into separate 2/3/4 phase modulators running at different rates? Or independent randomized (piecewise linear or whatever) modulators? Is it preferred to apply complementary modulation to consecutive APFs? There are a huge number of options, and I'm not sure I'm perceptive enough to find the "best".

3. I'm not a huge fan of FIR "early reflections", but they do come in handy sometimes. I've found they're quite hard to make from scratch and generating them randomly just results in 10000 flavors of crap. I've gotten a few from papers and such, but is there any sort of repository or some that people wouldn't mind sharing?

3.14159... Speaking of which, Moorer notes that an APF with a non-allpass filter in the loop can be made allpass if preceded by its time reverse. It seems like a multitap FIR type thing used this way could take care of early reflections and diffusion all at once- anyone tried it?

4. Regarding multi-band reverbs, is there much advantage here besides independent control over reverb times? I know Yamaha did this a lot, but did anyone else? Do different delay times for different bands result in a smoother sound?

Christ, I've nearly written a doctoral thesis myself...

[Froombosch]

Hello Acreil,

Thats a lenghty post. There are a few papers and patents in the effectprocessor forum and some older discussions on building algos. Feel free to add more if you have found some.

My work is in the constant density reverb type, so most of your questions are for other people around here. I am not a big fan of allpass filters. I am working now with inversed feedback combfilters. The darkening sound of these filters sound more natural to me.

[acreil]

I'll probably be doing a constant density thing sometime soon, some sort of extended Moorer/Ursa Major 8x32 type algorithm with some EMT-inspired enhancements (and whatever else I can think of). Even with all the design metrics that have come up over the years, they still seem more difficult to tune than the "big loop with a bunch of APFs" topologies.

The best hardware reverb I have at the moment is an Ibanez SDR-1000, which was very well received in this thread (anyone got the "+" OS ROMs?). While there's no modulation, it's a quite nice sounding constant density reverb, so I'm not really worried about making some sort of gold standard.

Part of what got me curious about reverb is pondering what makes vintage boxes so desirable, and if there was some cheap and overlooked shortcut to the "vintage sound"- turns out the SDR-1000 was a pretty good choice. I'm still perpetually intrigued by some other "underdog" boxes, the Dynacord DRP-20, Korg DRV-3000, etc. but if my synth buying habits have taught me anything, it's that obscure and overlooked items, while sometimes very interesting, aren't always the most useful.

[seancostello]

Originally Posted by acreil

I'll probably be doing a constant density thing sometime soon, some sort of extended Moorer/Ursa Major 8x32 type algorithm with some EMT-inspired enhancements (and whatever else I can think of). Even with all the design metrics that have come up over the years, they still seem more difficult to tune than the "big loop with a bunch of APFs" topologies.

I think that this is the trickiest part of reverb design: tuning the delays, and knowing how to tune them for a given topology. The allpass loops are fairly forgiving; feedback delay networks / parallel combs less so; FDNs / parallel combs with lots of output taps even less so. Once you start embedding allpasses within FDNs, it gets ridiculous. I spent the first half of the 2000's learning how to create stable structures, and the second half learning how to make those sound good.

As far as papers, if you have a local university, try going there and see if they have subscriptions to the journals in question. I head over to the University of Washington all the time, and email papers to myself.

[acreil]

Originally Posted by seancostello

I think that this is the trickiest part of reverb design: tuning the delays, and knowing how to tune them for a given topology.

That certainly seems to be the consensus. Some of the (obviously more encouraging) topologies I've tried sound almost acceptable just on the first "mash the number pad" test. It seems related to the total delay time in the loop(s), ignoring the APFs, which I guess isn't surprising. I saw your posts with Keith Barr on the Spin Semiconductor forum, and I think his "one big loop" topology really has a lot of advantages that way. When I built it I didn't have any context and didn't expect much (it's a tutorial, after all), but it was immediately surprising. Multiple parallel loops seem to "fight against each other" a lot more. Tuning a single loop algorithm feels more like "enhancing and improving" rather than "making it less obviously bad".

I wonder if the constant density parallel comb filter case can be improved by using fewer combs and adding more output taps per delay... I suspect it's really only trading one problem for another...

[seancostello]

Originally Posted by acreil

Multiple parallel loops seem to "fight against each other" a lot more. Tuning a single loop algorithm feels more like "enhancing and improving" rather than "making it less obviously bad".

The thing that is nice about a single loop is that the output taps will always have the same relationship to each other. Make those sound nice, and you're in business. The allpasses within the loop can be viewed as adding phase randomization to the loop, although they can also cause artifacts if not tuned properly.

I wonder if the constant density parallel comb filter case can be improved by using fewer combs and adding more output taps per delay... I suspect it's really only trading one problem for another...

I look at each comb as a wheel, and the taps as spokes. You have wheels rotating at different rates, and they are on top of each other. Each time the spokes of one or more wheels are at the same place, an unpleasant resonance is formed. So...is it easier to reduce the overlap with a bunch of wheels with less spokes, or less wheels and more spokes?

[madtheory]

acreil, I think the Ibanez became the Sony DPS box, there's a Roger Nichols article that refers to that. Still a very nice sound.

[Froombosch]

Originally Posted by seancostello

I look at each comb as a wheel, and the taps as spokes. You have wheels rotating at different rates, and they are on top of each other. Each time the spokes of one or more wheels are at the same place, an unpleasant resonance is formed. So...is it easier to reduce the overlap with a bunch of wheels with less spokes, or less wheels and more spokes?

One wheel would get a more evently distributed spoke density, but would repeat itself. Repeating can be seen/heared. It is not easy to build a good distribution with two, three or more wheels, but why take the easy route??

[seancostello]

Originally Posted by Froombosch

One wheel would get a more evently distributed spoke density, but would repeat itself. Repeating can be seen/heared. It is not easy to build a good distribution with two, three or more wheels, but why take the easy route??

Repetition can be heard with 2, 3, or 20 "wheels" (tapped combs) as well. For time-invariant reverbs, the best you can do is to have a repetition rate that is the sum of all of the delay lengths. 16 parallel combs with a total delay of 500 msec and a single delay line with lots of taps and embedded allpasses with a total length of 500 msec will both have a repetition artifact of 500 msec.

[acreil]

Originally Posted by seancostello

The thing that is nice about a single loop is that the output taps will always have the same relationship to each other. Make those sound nice, and you're in business. The allpasses within the loop can be viewed as adding phase randomization to the loop, although they can also cause artifacts if not tuned properly.

I look at each comb as a wheel, and the taps as spokes. You have wheels rotating at different rates, and they are on top of each other. Each time the spokes of one or more wheels are at the same place, an unpleasant resonance is formed. So...is it easier to reduce the overlap with a bunch of wheels with less spokes, or less wheels and more spokes?

The thing I'm sorta wondering is whether it would be worthwhile to make a "multitap delay with single feedback path" where the multitap part is fairly dense. Of course this will be annoyingly periodic, but maybe if several of these were used in parallel it would reduce the easily perceived periodicity without (er, hopefully) making it much harder to tune. Maybe a single very dense FIR could be "decomposed" by alternately assigning taps to different loops, such that the initial output (before the input signal reaches the feedback taps) would be identical to the original impulse response. Then, uh, put a modulated APF on the feedback tap and hope for the best.

What got me thinking along these lines is that the "single loop" topology basically reduces to a multitap with feedback if all the APF coefficients are zero. This initially didn't strike me as a great idea, but the results are good.

[acreil]

This is an example of my random delay based reverb.

There are 8 delays per channel, each with crossfading random delay taps (separate "audition" and "feedback" taps), followed by 2 APFs with both piecewise continuous random and sinusoidal modulation. There are 3 feedback paths- a delay can feed back into its own input, into the next delay's input, or groups of 4 delays can be mixed and fed into the 4 delays on the other channel.

Don't take the "granular" name too seriously, I don't think it really has anything to do with granular synthesis as described by Barry Truax, it just needed a name.

Obviously this is an extreme example with less than compelling source material (crappy Yamaha TG77 marimba sample), but I was rather surprised that the tail does smooth out without turning into a mess. Somehow a bunch of apparent no-nos combined to form something useful (to me, at least). It can be toned down but probably not to the extent that I'd consider it a general purpose style effect.

[zmix]

Originally Posted by acreil

...

The best hardware reverb I have at the moment is an Ibanez SDR-1000, which was very well received in this thread (anyone got the "+" OS ROMs?). While there's no modulation, it's a quite nice sounding constant density reverb, so I'm not really worried about making some sort of gold standard.

Part of what got me curious about reverb is pondering what makes vintage boxes so desirable, and if there was some cheap and overlooked shortcut to the "vintage sound"- turns out the SDR-1000 was a pretty good choice. I'm still perpetually intrigued by some other "underdog" boxes, the Dynacord DRP-20, Korg DRV-3000, etc. but if my synth buying habits have taught me anything, it's that obscure and overlooked items, while sometimes very interesting, aren't always the most useful.

Don't be fooled. The Ibanez SDR-1000 was designed by Sony. They put thousands of man-hours into researching reverb, and their 'first' attempt, the 1979-80 era DRE-2000 is a work of art. The SDR-1000 was comissioned by Ibanez, and represents the next step: it utilized a dedicated VLSI reverb processor (Sony called it the "Presence Chip"). After a falling out with Ibanez, Sony released their own version of the SDR-1000 and called it the MUR-201. Both the SDR-1000 and the MUR-201 are "True Stereo" devices, too.

[volumetric]

Hello Everyone. Something you may want to look at -

Dattorro - Wikimization
Talk:Dattorro - Wikimization


hmmmmmmm....

[zmix]

Flux:: sound and picture development

Flux:: sound and picture development

IRCAM SPAT - The complete Room Acoustics Simulation and Localisation Solution

With more than a decade of research performed by the Acoustic and Cognitive Spaces Team at IRCAM, being at the forefront of scientific and technological innovations, the SPAT is the most advanced and sophisticated tool for room acoustics simulation and localisation ever designed, managing both spatialisation (source localisation) and room acoustic simulation in a truly consistent and visually logical way.



Designed for surround and multi-channel use, SPAT presents the option to setup the output arrangements providing a variety of stereo and surround configurations, including subwoofer configuration. With eight input and output channels available in SPAT, configurations up to 7.1 and 8.0 are feasible.



Each of the up to eight incoming audio channels is internally mapped to a range of Virtual Sources localized in a 3D space, and connected to a room (a reverb). Up to 3 rooms in parallel are provided, presenting the option to simulate complex spaces (coupled room acoustics).



SPAT introduce state of the art techniques for room acoustics simulation utilizing advanced perceptive models, concealing the complexity behind the actual algorithms, allowing for intuitive and accommodating user interaction capabilities.





The SPAT licence includes the IRCAM Verb as well.

[Casey]

Originally Posted by acreil

Part of what got me curious about reverb is pondering what makes vintage boxes so desirable, and if there was some cheap and overlooked shortcut to the "vintage sound"

I don't think there are any real shortcuts to reverb. Most everything is based on fairly simple topological reasoning and the associated math. Tweaking is really just a way of learning this, as it appears you are finding out!

If you think of convolution as a shortcut; Even there, Sony still has the best example of that technology.



-Casey

[acreil]

Mostly I had naively wondered if the vintage "discrete DSP" architecture had some speed advantage over a general purpose DSP. I was thinking of how, for example, the hardware implementation of FM in the Yamaha DX7 (and to a lesser extent, the Synergy) is far more efficient than software running on a DSP. But of course this is based on clever use of shift registers and avoiding multiplication, and doesn't really apply to reverb.

I think a better explanation is that the difficulty of the problem in the early years attracted a more dedicated class of designer that was willing to spend a great deal of time experimenting and refining the algorithms. They worked well ahead of any published research, without necessarily having a great deal of prior knowledge of artificial reverberation, or even much of a preconceived notion of what digital reverb is supposed to sound like.

That's extremely impressive, and more than a little humbling when I can get something basic happening in about an hour, but then I'm too lazy to tune it much better than "not completely awful".

Also, on a somewhat related note, are the Ursa Major products the only ones that used significant analog processing inside the actual reverb engine?