(How) Does a MN3007 alter the sound?

[Jayzon]

This is more a theoretical question, but i hope to gain better understanding our choices for designs and components this way.

When I see so many popular designs based on specific chips, I always wonder what's special about those chips. This is even more so, when people try to emulate or replicate older designs (nothing wrong with this). For example, some of my favorite pedals / effects units use MN3007  chips.

But my question is, in what way does a MN3007 BBD chip - possibly in combination with MN3101 - alter the sound going through it?

I know its a delay, and that the circuits surrounding it usually do some pre - postprocessing to suit its frequency response. But is that it? Does it only delay the sound, or is there some noticeable distortion or phasing or ... going on?

[teemuk]

It's a "bucket brigade" delay. The name comes from it's operating mechanism; Basically it's a circuit that stores a charge that corresponds to signal's amplitude at the time of sampling into a capacitor. At a certain clock frequency the clock pulse-controlled internal switching mechanism passes that charge to another capacitor while the former capacitor get's recharged - this time with new information corresponding to input signal's amplitude at this new moment of sampling. Thus the sampled informations are carried along this delay line like and old-school fire brigade used to pass on buckets of water from one person to another. The lenght and operating speed of the delay line determine the delay time. Usually the BBD has several hundreds of these delay stages.

So, as it essentially is a sampling and switching circuit the sampling frequency limits the highest frequency the device can pass, as determined by Nyquist-Shannon sampling theorem. The switching also introduces plenty of noise.

Most of the frequency shaping is to compensate those effects: High frequencies need to be limited pre-BBD to minimize amount of "aliasing" effects and distortion born from trying to sample a signal frequency that exceeds the BBD's capbilities. Post BBD the signal is also low-pass filtered both to eliminate switching noise and to "reconstruct" the signal that is basically built from DC pulses having a particular amplitude.

Why people choose certain BBD in oppose to another... well, my guess for the various reasons are
- The chips performance, like how many delay stages it has an what is the clock frequency
- Availability
- Pre-establishment of tried and tested designs
- Mojo. People choose a device because they simply believe it has some magic properties. Moore people choosing to use the device bacause of same reason has an piling-up effect and soon people universally think the BBD has some special qualities that other ones are lacking.  

http://www.synthdiy.com/files/2003/MN3004+MN3011+MN3101.pdf

[Jayzon]

Thanks Teemuk for your clear answer. I knew a bit about BBD's but this helps me even further!
The function of the chip, and the pre-filtering is very clear now.

The only thing i am still wondering, if is the bucket delay mechanism itself (in between bucket #1 and #512/1024), has a noticable influence on the sound.

[merlinb]

Thanks Teemuk for your clear answer. I knew a bit about BBD's but this helps me even further!
The function of the chip, and the pre-filtering is very clear now.

The only thing i am still wondering, if is the bucket delay mechanism itself (in between bucket #1 and #512/1024), has a noticable influence on the sound.

If you look at the datasheet
http://pdf1.alldatasheet.com/datasheet-pdf/view/14235/PANASONIC/MN3007.html
page '36' shows a graph of distortion versus level. You can see that if you exceed 1Vrms then it starts to skyrocket (I also wonder how accurate this graph is..) Better than that though, check out the THD v. Bias graph. Unless you get it spot on you will get seriously noticeable distortion! I don't know how the classic pedals measure up though, but it's not unreasonable to suggest that that chip is "of its time" (meaning, it distorts, and you can hear the change!)

On the other hand, maybe all the extra support circuitry in the pedals was done on the cheap, so THAT made more of a difference? I see the example schem in the datasheet seems to show some serious high-cut filtering, even though the chip claims a bandwidth exceeding 12kHz...

[Fender3D]

Your signal samples will feed (switched) capacitors in your BBD.
The caps chain in the BBD will slowly discharge as time goes by just like every capacitor does (those damn high voltage aside  ... ) thus your output waveform will differ from the input one depending on sampling rate/frequency/level.
This is not the usual kind of distortion...

Usually the Reticon's BBDs have a higher output signal than Matsushita's, thus obtaining a better overall S/N ratio.

[PRR]

> some of my favorite pedals / effects units use MN3007  chips.

ALL your older BBD effects use the "SAME" chips.

There's just one basic design, patented, and not made at more than one or two companies.

As time went on, longer (more stages) versions became possible and affordable, from 256 to 4096. Also there was some improvement in distortion (essential to making long delays tolerable).

We used it because that was the ONLY $20 way to get a significant short delay. All-pass networks are too short for many fun tricks, tape-delay costs a lot more, and ADC-RAM-DAC digital techniques were a million bucks when the BBDs came out.

All BBDs distort. Intensely when the signal is anywhere near the power rails (+ or -). Less if you stay in the "center". For a small range around "center" the THD can be low enough to look OK. But "center" varies from part to part so you usually want to trim. Bring a signal up until it sounds bent, trim the trimmer for minimum bentness, repeat with a larger signal. That will do for rock and roll. For "hi-fi" (hah!), then drop the signal somewhat below the bent-zone, and trim for purest sound (or lowest THD). Generally the setting for largest not-too-bent signal is not the same as for purest small signals.... your choice which way to go (you may not really care).

There's new-made BBDs from CoolAudio but I believe they are the same circuits re-laid-out on new smaller Silicon. I think it is intended to work-alike. Any re-lay-out of course changes small details.

[12Bass]

On the other hand, maybe all the extra support circuitry in the pedals was done on the cheap, so THAT made more of a difference? I see the example schem in the datasheet seems to show some serious high-cut filtering, even though the chip claims a bandwidth exceeding 12kHz...

The cutoff frequency of the low-pass filter depends upon the clock rate of the BBD.  For high clock speeds which are well above the audible range (like in flangers), the low pass filter can be moved to a higher frequency to take advantage of the full bandwidth of the BBD.  However, in something like an echo/delay, often the clock rate is near audio frequencies, requiring a reduction in the bandwidth of the signal going through the BBD in order to avoid aliasing.

[daverdave]

Yeah, as far as I've read the lowpass filter cutoff should be 1/3 of the lowest clock speed. Typically delay pedals use companding to avoid noise or distortion from the bbd, but I haven't seen many flangers or chorus pedals that use it.

There's a good book called 'Audio IC User's Handbook' by R M Marston that covers both analogue and digital delay lines. It's got a bunch of circuit examples for different BBDs and clock circuits, as well as lowpass filters. It also covers the NE570 compander ic, but for some reason doesn't mention using them in the delay lines. It's well worth a read though if you want to know a bit about BBDs.

[Mark Hammer]

In general, it is not the BBD that alters the sound so much as all the things you have to have WITH the BBD to make it do what you need it to do.  So, much of the blathering on about "analog-delay sound" is really more about the filtering needed to keep the clock noise and aliasing out of the signal.  Tack that same filtering onto a PT2399 and you instantly get a lot of what people are talking about when they talk about analog delays.

I suppose there are some potential audible differences between delay chips (e.g., the MN3002, used in the Boss CE-1, or the differences between MN3005 and 3205, etc., but personally I've never heard them).

In general, though, because of the nature of bucket brigades, the quality of the sound will be inextricably linked to the manner in which they are stepped/clocked through their duties.  faster clocking will not only result in more snapshots per second (i.e., a higher sample rate, and more faithful representation of the entire signal), but will result in less signal loss per stage via cap leakage.  Note as well that whatever signal leakage occurs is magnified by recirculation.  If I'm clocking a 4096-stage device slowly and feed some of that signal back to the beginning, it has 4096 chances to leak yet one more time.  Make that 5 repeats at maximum delay, and you can see why the signal degrades over repeats, and why the filtering has to be so severe in order to make the audio quality seem halfway respectable.

The solution, though, is not always to clock as fast as possible.  EVERY BBD has a certain amount of capacitance on the clock input pins, that acts to "un-squarify" the clock pulse if that clock pulse is too fast.  My understanding of it is that a properly clocked BBD will output effectively contiguous analog samples (i.e., whatever gap there might be between them is so infinitessimally small you won't hear it).  At a certain point, however, once the clock starts to exceed frequency X and is not suitably buffered to provide adequate current, the square quality of the clock pulse turns into a McDonald's arch, leaving larger, and most importantly audible, gaps between successive samples at the output.  Imagine taking out every 3rd or 4th frame of a film and replacing it with a blank, and you'll have some sense of what the audio impact might be.  This upper limit for unbuffered clocks is why the datasheets for the Panasonic chips seem to indicate 100khz as the absolute maximum clock rate; they assume you are necessarily using an MN3102 or 3101 clock to drive the BBD directly.

[Labaris]

What a great post!! Thanks for the info

[StephenGiles]

"they assume you are necessarily using an MN3102 or 3101 clock to drive the BBD directly".

Quite right and I have always used 4047s for clocks - because I can!

[Labaris]

"they assume you are necessarily using an MN3102 or 3101 clock to drive the BBD directly".

Quite right and I have always used 4047s for clocks - because I can!

Are MN3102 or 3101 too expensive?
What's the reason for not using them?

[oldschoolanalog]

Are MN3102 or 3101 too expensive?

Compared to a 4047 yes. Relatively speaking no.

What's the reason for not using them?

See the last sentence in Mr. Hammer's post above.
A 4047 can be easily clocked up to ~2MHz (with proper buffering of the clock lines) IIRC.

[Mark Hammer]

The difference between using the MN3101/3102 as clocks, and other types of clocks is that the clock pulse is adequately buffered in the 4046 and 4047 to overcome the input capacitance of the BBD.  Look at the Hollis Ultraflanger or the Anderton Hyperflange, and you'll see that sometimes a 4049 or similar CMOS chip is used to buffer the clock lines.  Again, it is not that the 3101/3102 are not capable of providing outputs greater than 100khz.  Rather they don't provide the sort of outputs that work well above that frequency if the clock lines are tied directly to the BBD clock inputs.  In that sense it's like the difference between running a 600ohm voice mic over a 50ft cable, and running a guitar pickup over the very same cable.  The mic signal is not impacted quite as much by the long cable as the guitar is, because of their electronic (chiefly impedance) differences.  Run the mic and guitar through a 2ft cable and you won't notice any difference in their fidelity; it's what challenges fidelity that you concern yourself with.

Now, why use an MN3101/3102 if they have those sorts of shortcomings (and cost more than CMOS chips)?

1) You won't find a smaller package to do the job, nor a simpler solution to provide the required function.
2) If you want to keep clock noise and bleedthrough to a minimum, it's good to keep the lines as short as possible, and  it doesn't get much shorter than snuggling an MN3101 right alongside an MN3007.
3) The high-frequency limitations of the 3101/3102 really only matter when you are aiming for the shortest possible delay.  Most of the time, the things we want/need BBDs to do for us fall well within the 100khz upper limit.  I mean, you could overclock two MN3005s in series for the highest resolution chorus known to humankind, but why bother when a normally-clocked MN3007 does the job perfectly acceptably?  And who would want to overclock an octet of MN3008s to get a maximum delay of 200msec?  So, the only time converting to a CMOS clock generator really starts to matter is when you're aiming for cheap and/or short delays.  You CAN achieve shorter delays, of course, by means of low-capacity BBDs, but most would agree that you get more fidelity by overclocking a 1024-stage chip (for higher sampling rate) than by normally-clocking a 128-stage device to get the same amount of delay

[dmc777]

This has turned into a great post. Very useful information in here. I'm gonna have to read a dozen times or so to completely understand it all but you guys are really doing a great job at explaining this. Tell me more!