A stupid question concerning delays

[Thomas P.]

I should know that this has been dicussed over and over again but I believe I missed it :?

How the story starts:
I looked at the new stocklist of my No.1 parts supplier and again came across the MN3008 and MN3101 which are available at a very good price.
My thoughts were that there has to be something good about it and I started to search for projects.
The first (and proppably the easiest) was on the datasheet called a 'reverberation effect'. I studied it for a while and saw that the MN3008 is designed for a negativ voltage as all the chips from this series.
At the archive I found that someone had used a MN3008 in the zombie. Since I've already got a zombie I'm really into building a reverb etc.
What I did was searching for delay projects and found that some are makeing a diference between delay and echo which I ever thought has to be almost the same.
With this in mind I looked at other datasheets of the MN300x series and on everyone was the same 'reverberation' circuit once called like this once called 'echo' :?
But help was on its way: I found a site with some examples (also concerning electronical designs). The first note was that reverbs designed on BBD are very poor. And further on there was also this difference between delay and echo.

So my question is: What is the (technical) difference between them?
Am I right that each one delays the signal putting some of it back into the delayline and on it's output mixing it with the dry signal?

[ExpAnonColin]

As far as my experience goes, the basic sort of "unwritten but agreeed upon" difference is that delay doesn't self oscillate and echo may.  You are right in saying that both of them involve delaying the signal in a delayline and then mixing it with the dry signal (but you're missing the feedback stage where the 2nd, 3rd, 4th, etc echos come from).  I find that it's pretty safe to say that delay's don't oscillate and echo's do.  Think about a DL4, the "delay" settings are basically the ones that don't self oscillate.  The DE7 has 2 modes, one for delay and one for echo, the main difference is that the echo self oscillates.  Other than that, they work very similiarly for the most part, and often it's up to the builder whether or not he wants to call it a delay or an echo.  Although I suppose it could be said that a flanger, a chorus, a reverb, and an echo are all delay effects-because they are all based on delaying the signal, but I find it's safer to go with the other way.  So it's one of the two, either "delay" meaning it doesnt self oscillate or "delay" referring to the fact that the audio signal is delayed and not explaining the exact workings of the circuit.

-Colin

[Peter Snowberg]

Delay: Input is delayed by XX milliseconds

Echo: Input is delayed by XX milliseconds and then some is fed back into the input to create repeating

Reverb: This is a much more complex effect involving many parallel echos or similar topologies

In each case the dry signal is (usually) mixed with the delayed signal. Echo is just delay with feedback added. BBDs make poor reverbs because reverb effects rely on lots of parallel delay lines and lots of filtering. It's not that BBDs are by default poor components for a reverb, but to make a good reverb out of them you would need MANY BBD chips and the noise would overload everything unless the design was very careful. It would be expensive, huge, and inflexible. DSPs are much better components to make reverbs out of (but, springs are still my favorite ).

For more information about reverbs, look at this paper:
http://www.stanford.edu/~dattorro/EffectDesignPart1.pdf
from
http://www.stanford.edu/~dattorro/ who used to work for Lexicon and Ensoniq. The publication of this paper in "Journal of the AES" resulted in lawsuits and his web page is now the only place to find it outside of university libraries. Be sure to read the other two papers in that series if you are interested in DSP.

Take care,
-Peter

[Mark Hammer]

Stand in a room, any room, and make a noise.  What you hear is the sum of all reflections bouncing off of everything in the room; sometimes after it has bounced off several surfaces before arriving at your ear.  

Reverberation in natural spaces is a more diffuse sound than  a simple, straightforward echoic repetition.  In very LARGE defined spaces with very specific sorts of reflecting surfaces or dimensions, you can get specific reflections standing out from amidst everything else, and that's what echo is.  But in nature, such echoes usually occur against a backdrop of other diffuse reflections.

The other thing to remember is that, even in ceramic tiled bathrooms or cinder-block gymnasiums, reflected sound does not have the same bandwidth as the sound source, and certainly does not have the same bandwidth over time.  Each time sound has to bounce off a surface and travel even farther through the air, high frequency content suffers most and gets lost.

Springs and large thin metal plates have traditionally been used to mimic real reverberation because they have mechanical properties similar to natural spaces.  That is, they provide multiple "replicas" of the original sound at different delay times, and also lose bandwidth quickly.  Indeed, if you have an amp which allows you to isolate the reverb signal (an old solid state Gibson I used to have allowed you to hear reverb only), you'll hear something that has only a vague resemblance to the original because of the way that restricted bandwidth and multiple reflections obscure the original.

What gets *called* "reverberation" in appnotes schematics is nothing of the sort, and is more properly described as discrete echoes.  Again, echoes are found in certain *types* of reverberant spaces, so it shouldn't surprise us that adding a bit of echo makes a sound seem to come from a defined space, and varying the amount of delay can make that space seem larger or smaller.  But realistically, the relationship between BBD-based echoes and "real reverb" is very much like the relationship between a black and white line/vector drawing that tries to convey 3-d depth through use of perspective, overlap, relative size, etc., and a fully-rendered, texture-mapped, and shaded, 24-bit colour image of the same 3-d space.  There is barebones 3-d depth and realistic 3-d depth, and there is simulation of ambient space and recreation of ambient space.

The closest that BBD chips ever got to producing something like reverb was the multi-tapped chips like the MN3011 (6 taps) and MN3214 (5 taps).  The intent was for each of the taps to be mixed down at the output and have a simulation of multiple reflections without standing waves or undue resonances (what you get if you set short delays and crank up the regen - the "boxey" sound).  While better, these chips were never really fully exploited and, truth be told, you probably needed a couple of them and some well-planned tap-specific filtering to fake reverb anyways, so they never caught hold firmly enough before digital chips made better reverb emulation cheap enough.

Personally, I think you can probably emulate reverb by using a couple of cascaded BBDs set to different clock rates, using combinations of feed-forward and feed-back to mixing stages, but that starts to get complicated and costly with respect to pots, knobs, package size, etc.  A well-designed spring system (including pre and post EQ) probably fakes it better and cheaper, and easier.

[Joep]

Colin,

What do you mean with (self) oscillation in you answer? Feeding back the delayed signal?

Thanks,

Joep

[puretube]

in fact: feeding "too much" back, till you hear a metallic "eek"...

[Samuel]

I think, with respect to self-oscillation, he means that the signal repeats and builds upon itself, to the point where, without additional signal input from the guitar (or whatever source), the effect will continuously create output, because a loop is formed. The ouput is fed back to the input with little or no change in level, so the circuit is constantly creating output regardless of external input.

This doesn't necessarily mean unwanted noise, like squealing or 'feedback' in the Jimi Hendrix holding the headstock of the guitar against the amp kind of way. Looping pedals, like the PT-80 at full delay repeats setting, or the Boss Loop Station would be an example of an intentional "self-oscillating" system, although the specific term "self-oscillation" usually (in my experience) refers to a nasty byproduct of high gain circuits which can reamplify their own internal noise and begin to shriek regardless of input from any source, particularly if the input and output wirings are brought too close to one another...again this can be intentional of course, just check out colin-anonymous' Parallel Universe, or troubletom's Noise Swash for circuits that intentionally loop output to input to get this effect.

[Thomas P.]

What gets *called* "reverberation" in appnotes schematics is nothing of the sort, and is more properly described as discrete echoes.  Again, echoes are found in certain *types* of reverberant spaces, so it shouldn't surprise us that adding a bit of echo makes a sound seem to come from a defined space, and varying the amount of delay can make that space seem larger or smaller.  But realistically, the relationship between BBD-based echoes and "real reverb" is very much like the relationship between a black and white...

I thought that it would be an echo but I can't see where the delayed signal is fed back to the input...

[Mark Hammer]

I'm looking at the datasheet for the MN3008 that I have in the BBDementia.zip file at my site.  On the last page of the datasheet is a schematic for a "reverberation effect generation circuit".  In that schematic, IC2 in the lower right hand corner takes a feed from the output stage, via an "echo control" pot and feeds it back to the input stage.

[Thomas P.]

How do you think will it sound? Strange question, I know but I'm not sure if it's worth the effort.

Would it be possible (for that particular circuit) to cascade two MN3008? I know that the MN3101 can handle this amount of stages but how do I connect them together?

[Mark Hammer]

The MN3101 is spec'd to be able to clock up to 8192 stages (equivalent to four MN3008's), although I suspect that spec assumes no additional buffering of clock line signals.  The clock inputs on the BBD have a certain amount of capacitance, and once you start paralleling that across multiple chips, the square clock signal starts to turn to mush (input capacitance is one of the reasons why you can't clock MN3007/3207's into sub half-millisecond delays with the standard MN3101 unless you use "outside help"; the clock signal doesn't hold it's "squareness" well at high speeds without additional buffering).

How will it sound?  It  will sound like basically any other analog delay line, different only in terms of what cutoff frequency the anti-alias filters are set to.

Do remember that a great deal of what gets described as "analog warmth" of such delays in promotional literature is a load of poppy%^&*.  The"warmth" people associate with them is a result of the severe filtering needed to keep the clock whine out of the output.  In this case, "warmth" = restricted bandwidth.  Digital attempts to do the same thing will seem less "warm" because the technology does not require clock speeds low enough to require that sort of filtering.  As a result, you get more bandwidth and a "less warm" sound.  Of course, thereis nothing to stop anyone from making the filters roll off at a lower requency so as to achieve "warmth".

To see how to hook them together, take a look at the delay project at generalguitargadgets that uses a pair (or is it 3?) of BL3208 chips.  The basic logic is that you mix the two outputs of the chip together, stick in a DC blocking cap, and rebias at the input of the next chip.  I was always under the impression that chip biasing was something you had to do on a BBD-by-BBD basis in a multiple chip system, but apparently a single bias trimpot can successfully feed a couple of chips with little problem.

[Thomas P.]

Thanks Mark, and of course all others!!!

I think I will try this useing two MN3008. I looked at the AD-3208 project and it is like I thought. But in this project every BBD input is biased with trimmer. How is it in this application circuit? Does the Ic provide a bias voltage for the BBD input?

[Mark Hammer]

There is no requirement to bias with only a single trimpot, but apparently there is no requirement to bias with separate ones either.  I always thought you needed the precision that separate trimpots provided, but others have assured me that a single trimpot feeding two BBDs will work just fine.