Reverb designs

Hi,

I use Scope DP mainly for testing different types of reverb algorithms, including time-variant FDN reverberators. So ask away, if you need help

Just a sanity check... FDN is basically a many-tap delay right or are the taps modulated in some weird way?

Hi Guy,

FDN stands for Feedback Delay Networks. FDN can be represented as a parallel tank of taps interconnected via a energy-preserving feedback matrix. This matrix can be modulated in different ways.

Apparently, allmost all reverb plugins for Scope/Pulsar use a Moore/Schroeder algorithm. It's based on a tank of parallel comb filters, which is connected to a tank of cascading allpass filters. This approach have some drawbacks : Tonal coloration of the tail and lack of high echo density - They sound Metallic.

Do you know how Quantec do it? i know they do things a bit different.

On 2001-06-19 07:48, Martin Lind wrote:
BTW, how can I include pictures??

just include HTML in your post. in other words you'd write a link like <IMG SRC="link-to-your-picture">

cool discussion btw
-john

Nope, I have never tested a Quantec reverb. But my guess is they use a combination of UFDN (allpass feedback structure) and nested allpass filters...... Hmm I don't know - sorry!!!

Hi,

Here are some pictures explaining the Moore/Schroeder reverb algoritm :

1) The two tanks (Comp filter and Allpass filter)

<IMG SRC="http://maeli.punqtured.dk/Scope/Tanks.jpg">

2) Inside the Comp Filter Tank - as you can see, there are 8 comp filters. This is mono, if you need stereo, you have to add 19-35 to all the delay values in the right channel.

<IMG SRC="http://maeli.punqtured.dk/Scope/Comp Filter Tank.jpg">

3) Inside the Allpass Filter Tank - As above, if you need stereo.

<IMG SRC="http://maeli.punqtured.dk/Scope/Allpass Tank.jpg">

4) And now we are inside the Comp filter.

<IMG SRC="http://maeli.punqtured.dk/Scope/Comp Filter.jpg">

I hope you all understand what I mean, if not - Don't hesitate to ask!!!

Hhmm,

you can download the pictures from :

http://www.filesanywhere.com
login : Scope
pass : pass

um, yeah, i should have mentioned that when you do the img src = "link" - that link must be accessible on the internet! (not a link to C:\ on your hard disk!)

btw, you can edit your own posts to fix mistakes (just click on the edit button underneath your post).

-john

http://www-ccrma.stanford.edu/~jos/cfdn ... works.html

here's a link for FDN's and waveguides

Great Post Martin,
I would love to talk reverbs with you, my new versions use something similar to this,but at the same time still use the Moore/Shroeder design I buit the feature into the devices ;0)
As well as some cool binaural functions which are working very nicely.

Cheers
Paul
EarlyFirst
http://www.earlyfirst.com

Hi Paul,

I have been working on a Reverb for some time now. Im using UFDN & 12 taps time-variant FDN. Just included early reflection.

Remember to modulate the taps in your reverb. You can use 1st order linear interpolation (or allpass interpolation for better sound quality). And don't be afraid to use things like chorus and compressor in your designs (just look at some Lexicon/TC Electronics hardware reverbs).

You are welcome to email me or just ask me here, if you have any quenstions

Best regards

What a post! Since beginning reverb was THE headache of the platform. PCM91? REALLY?? Dear Martin, I hope you do not plan to use this for exclusive own private use. (Can you imagine the hungry Pulsarians sitting before and reading this post? Screaming for reverb with last effort in their broken voices?) Please let us know when, where and how much. I had a look in the formulas. Puh! Not easy! So Pulsarians, let us smoke+drink a few less and put it into the reverb-savings-account!
Happy pulsaring
Micha

At the moment it's a private project (it's far from finished). The following things are still missing :

* A nice looking user interface
* Different types of early reflections (I only have one)
* Optimization of the algorithms
* Things I can't remember right now

Cheers

Thanks to Martin, I did a search on some key words (Jean-Marc Jot, Moorer, Schroeder) and came up with a whole list of papers and circuit diagrams on reverbs. Here is a quote which seems to capture the essence of a reverb:

Analysis and synthesis of room reverberation based on a.. - Jean-Marc Jot.. (Correct)
.... model Moorer noted the auditive resemblance between a concert hall impulse response and a white noise multiplied by an exponentially decaying envelope, and reported that such a synthetic response can produce, by convolution with anechoic signals, a natural sounding reverberation effect [19]. To obtain a frequency dependent reverberation time, he suggested using a filter bank and summing the subband signals after multiplying them with different exponential envelopes (Fig. 1) Polack [2] developed a time domain model complementing Schroeder s frequency domain model. In this model, a ....

http://citeseer.nj.nec.com/context/209676/0

Unfortunately for me, and I claim to be no expert, I haven't liked any reverb I've come accross since I heard the top of the line Lexicon at a demo (about 10,000 pounds. My mind doesn't work with figures that big very well, so excuse me if I'm out a couple of thousand). The demonstration team demonstrated the smoothness a reverb added to the sound of a snare drum, while only demonstrated voice through the lower end models. Well the question was begging... what did the drums sound like through the lower end reverb, so I asked if they'd kindly let me hear the drums through the latter: in my mind the answer was, 'why would you want to... it sounds so bad in comparison'.

I hoped that something on the creamware platform might save me from my expectations, but my experience so far is that the platform just isn't capabable of it.

BTW, I don't like the creamware delays or reverbs, they both sound harsh and metallic to me. For some reason metallic reverb kills my creative process. As for that excuse for a chorus! (Anyways, I can't criticise without a little praise... I do love the synths, routability, etc, even they if it did do my head comming to terms with them. I can but forgive).

On the reverb front, the software reverbs in Logic Audio have 'kind of' made up for this, and the sound of sonic timeworks reverbs (software) is 'okay'... not that I can justify a splash on the timeworks stuff yet. In general, the software reverbs fall down in that they sound 'thin' to me. There is something missing from them.

Am I alone, or is there one that perceives as I?

Jilhead

Hi Jilhead,

You're not alone - thats why I started this thread. I was hoping we could have a discussion, regarding 'high-end' reverb algorithms.

Cheers
Martin

Martin,
one hint when using the PC delays the feedback mdl is not needed.

Cheers
Paul

Hi Paul,

Yeah, I know. But I just thougt that the routing was more clear, when I included the feedback modul

Another hint, when you're working with delay lines, work in 48KHz. A delay of 1000 samples in 44.1KHz is only (1000*44100)/48000 = 918 samples.

Cheers
Martin

haha cool Martin,
That's what I thought but was making sure.

lately I have been testing at 96k
I got a nice little 96K *ONLY* verb coming PT 96K.
which means it was built tested and ran only at 96K in order for people to take advantage of this samplerate.



Cheers
Paul

Hi Paul (please check your PM),

I am using the array modul for the different samplerates, since it's pain in the *** when you are working with interpolations

Cheers
Martin

I've found this interesting discussion.

Just to say that 1 DSP will be filled with 4 delay lines of 1000 sample.

To create a large reverb you need at least 6 delay lines of >1800 sample (only for the comb filters) so it is impossible to use DSP data RAM to store ALL the relays without filling a scope system! (same problem with the All pass tanks)

We need to use the PC delays that strangely using the PCI to access to the RAM (I though that 32k was on board...

On 2001-06-24 23:40, Warp69 wrote:
Hi Paul,

Yeah, I know. But I just thougt that the routing was more clear, when I included the feedback modul

Another hint, when you're working with delay lines, work in 48KHz. A delay of 1000 samples in 44.1KHz is only (1000*44100)/48000 = 918 samples.

Cheers
Martin

I thought it could be interesting to explain here a little bit of the "Spread Engine" used in the DSPdev reverbs.

The Spread engine is included beetween the Combs and the All pass.

COmbs -> Spread Engine | AllPass | StereoSP.
|....................|
--- Taps Feedback ---|

In order to create a Leaving reberb several things are missed in the Moorer algo.
- when a sound bounce back from a wall, it does not goes back to the listener without bouncing off the other walls. That's why each Comb delay must be conserved by a feedback shame and re-injected in all the comb filters. This is what all the very knwown hardware manufacturer are doing, whatever the name of this algo is...
- BUT the sound of each Comb filter must be phased to include the domain frequency changes done by the delay (doppler).

The Spread engine splits each Comb signal in 6-8 little delays. These are not really delays they are changing in the time( like a variable delay line) and also the sound frequency changes like a pitch shifting.

The comb filter are transformed in a 2^Combfilters number of signal that are moving all the time in term of frequencies and level. That's create the density that lacks from most reverbs avalaible in plugins.

With the Verb Gold Still in development, you can tune the Spread engine to separate the Frequencies (more or less) on each CombFilter, and you can change each comb delay so you can see that how each Comb filter are transformed.

The Tap feedback is here to introduce the long feedback (like in a Canyon). In some large room, the signal will bounce like 1-2s before coming back to the listener.

This create an "echo" like effect, but it is hard to notice it because the feedback must be very low to avoid any feedback overload problems.

These were some "secrets", but like all the secrets they are very easy to program as soon as you know how it works... That's why no one like to talk about these...

What's wrong with the Timeworks 4080L?

I can't believe how good it sounds...

I have Lexicon PCM60, Alesis Midiverb II (incredibly underated and legendary reverb unit...)and various classic bits of reverb kit and think the Pulsar 4080L is awesome sounding...

far better than anything else software I've ever heard on any platform including Waves Rverb TDM...

Hi Nicolas,

Choukroun wrote:
-

when a sound bounce back from a wall, it does not goes back to the listener without bouncing off the other walls. That's why each Comb delay must be conserved by a feedback shame and re-injected in all the comb filters. This is what all the very knwown hardware manufacturer are doing, whatever the name of this algo is...

The name of that algorithm is FDN (Feedback delay network) and is not used in any of the 'high end' reverbs (afaik). They use UFDN, which is allmost the same algorithm - but do not include any comp filters. In most cases, its a allpass cross feedback with 2 additional delay lines.

For the best sounding reverbs you need time-varying delay lines and some interpolation, like 2nd-3rd order langrange -or allpass interpolation.

Cheers

Hi Guys,

Well don't forget in some cases the Allpass sections are *first* in the chain....


my secret of the day


Nickolus if you use modulated delays, they *are* on the dsp's only if you use the bigger versions will they need to piggyback the PC based delay......


Cheers
Paul

Hi Guys,

Another secret : Don't use comp filters at all

Modulated delays = Time varying delay lines??

If yes, are you using interpolation??

Cheers

Isn't there another parameter to add and that causes reverb modulation ? like a fine phaser. It doesn't sound like a phaser on verb, but may create nice subtele evolution.
I only saw this in EPS 16+ 44.1 reverbs (i guess it's on DP 4 too) and is something you could hear in a natural reverb (i've experienced large hall reverb everytime i went out home,for more than 20 year, the sound of this hall is a really cool reverb
However, i don't know if it happens in all reverb spaces (but it doesn't in most software verbs).
Also, how to modelize a cavern or underground lakes with stalamites

More seriously,how gated reverbs are made : i compared 2 of them that I own, namely the very old Boss RRV10, which has a really cool gated reverb, with guts and hard gate (the rest is crap compared to what you can have now).
On the other hand, the STW 4080L, which is great except for gate (my opinion): i guess it is because it seems to cut reverb when decay reaches a certain treshold, and that's a very different effect than the Boss one, which seems based on TIME rather than Treshold, so the Boss is very steady and not input-sound-volume dependant.
i don't know if that make sense though,
cheers

Hi,

Can you send me a wav with some examples?
I understand what you say with the *Time* not threshold thing but I need to hear it...


paul

Yes Paul, but please be patient i'm totally submerged with work during day and night.

the "non time" option is better explained as "gate" ("reverb cuts when its decay goes under a certain treshold", whereas the other option, as you understood, could be "reverb cuts after x seconds").

spacef
http://www.spacef.com

Is there a place when those pictures are
stored i would love to see more plz

Dear Folk, I've released the Verb GOLD.

I think that some explanation on how it works could be interesting for all, first there is a very bad picture of its design here http://www.dspdev.com/products_verbgold.asp

I think that COMB filters and ALL pass Filters tanks are both worth. They do not give the same result.

The GoldVerb do not uses any CombFilters. But AllPass needs to be encapsuled in a Feedback delay network otherwise you loose too much of you original signal dynamics.

The goal of the FDN is to keep the signal and re-inject it as much as possible in order to have the best density.

In the Verb GOld I've used 3 FND in parallel.
- one with 2 delays (use the Del1 & del2 button to play with them)
- one with the Vtube + a Waveshaper (the preset "Distorsion" boots the waveshaper for fun.
- one with several filters (Filter button) and a panning effect.

When I was studying the Lexicon (I've sampled almost all their presets), I've came to thses conclusions :
- they use a panning effect at the end of the chain.
- they use a flanger that they sync with the delay length to obtain this kind of doppler effect.
- they boost the bass and the treble.
- they use a waveshaper on the bass and on the treble to give the bass a "fire burning" like sound and to the treble the smooth that is not naturaly included in the original sound.

Even if I've not the same sound in the Verb Gold (it is not a copy and it was not the idea), the results can be very similar.

The difficulty with a FND is the feedback loop.
To have a very warm sound, you must maximize the feedback and stay at the limit of larsen. That's why each preset must be tunned internaly carefully.

Several other effects must be included in order to create more harmonics and more effects, flanger, and chorus are in play.

Finaly a Reverb is very close to a multi effect included in a feedback loop

There will be a Reverb available soon, that will make you forget all the other stuff...

STAY TUNED !!!

Yup on Monday the PT-FREE-96K will be released

I want to know what happened to the Lexicon PCM 91 killer. Was it ever finished?