sad1024 unique tone?

[nepalnt21]

does the reticon sad1024 (used as a singular 512 stage flanger bbd) have any unique tonal properties that couldn't come from another 512 stage bbd (like the mn3204)?

i really like the electric mistress and the ibanez fl303, they have really beautiful tone... is it simply the circuit? or is there some magical tonal mojo of those reticon chips?

(i read where mark hammer says it can be given higher frequency clock signal, iirc.)

i mean other than maybe getting shorter delays

[ElectricDruid]

If the distortion of the BBD is kept low, I don't see really how it *could* have much in the way of tonal characteristics, tbh. It's supposed to be the same signal you put in, only a little bit later!

There's a *lot* else in the famous flangers that can explain the tone, and yes, the short delay times is definitely a part of that. The rest is:

• The pre-delay filtering, or lack of it.
• The post-delay filtering, or lack of it
• The LFO sweep shape
• The clock circuit response to that LFO sweep (could be linear wrt frequency, linear wrt period, V/Oct, etc etc there are lots of potential options)
• Actual tone shaping in the circuit! This is particularly effective in the resonace/feedback path, but pre-emphais/de-emphasis is fairly comon too.

That's off the top of my head. There's probably more. As you see, the BBD itself is a small part of the equation.

[Kevin Mitchell]

Besides headroom and what Tom said (the circuit designed around the chip), the only tonal difference between devices that I can think of is dependent on how the BBD is used.

For example, there are only a couple one-chip solutions that could do something like parallel multiplexing - when you clock the two channels at opposite pulses to double the sample rate - like how they did in the ADA flanger (both the MN3010, SAD1024 and modern x2 MN3204 versions). However, if we look at a design like the DIY clone the Flintlock Flanger where an MN3007 is used, it's clocked at twice the speed which corrects the delay time and in turn gives twice the sample rate.

It is unique, but nothing we can't emulate with modern parts.

[Mark Hammer]

I agree wholeheartedly with both of these fine and knowledgeable gentlemen.

[12Bass]

Based on my A/DA Flanger clone, with minimal filtering, the SAD1024A seems to have more extended high frequency response than the Panasonic BBDs (well over 10KHz).  So, if using high clock rates, like in a flanger, it may have a more open sounding tone.  Run it slow, with lower frequency LPF that cuts out anything above 4KHz, it probably doesn't sound much different.

[Mark Hammer]

The high capacitance of the clock input pins on the Panasonic/Matsushita chips means that they don't readily support clock frequencies of over 100khz....when not supplemented by suitable buffering and current drive of the clock output.  The "corruption" of the clock pulse coming directly from an MN3101/3102 to an MN3204/3207/et.al. means that, as one approaches and goes past 100khz, the "handoff" from one BBD stage to the next is not smooth.  Actually, it's not absolutely instantaneous in the best of circumstances, AFAIK, but close enough that we don't hear it.

Use of proper clock buffering did not prevent the MN32xx-based A/DA flangers from being among the best and most musical ever made.  The Reticon chips, having a MUCH lower capacitance on the clock input pins, could easily be clocked superfast, without suffering sound quality.  MN32xx can also be clocked very fast, BUT one has to provide the proper buffering and current drive to get over that clock-frequency "wall" that divides mundane from exceptional flangers.

So, to answer the original question, Reticon chips won't intrinsically and necessarily have a unique or "better" tone.  It just takes more parts and effort to extract that same quality from Panasonic chips.  If a circuit design does not include that "extra mile", Reticons will likely sound better than Panasonics, in the flanger context.  Since choruses rarely clock that high, such differences won't generally show up in that context.

[12Bass]

Figure 7. on the SAD1024A data sheet shows frequency response nearing roughly 70kHz at 1MHz clock; however, I don't see the MN3007 data sheet going that high.  I know from experience that a SAD1024 running at 500kHz or higher in parallel-multiplex mode with relaxed filtering easily reproduces beyond 10kHz, as cymbals are surprisingly well resolved in the isolated delay path (the overall BBD path fidelity surprised me through AKG K702 headphones... a bit noisy, but impressively clear).  Also, bear in mind my bass rig is full range, going up past 15kHz, whilst most guitar amps roll-off the highs past 5kHz or so, meaning extended highs from the BBD path would be lost as well.  I spent quite a bit of time listening to music through my A/DA clone before it stopped working after getting knocked over.   

I've also listened to various comparisons of EHX flangers and choruses (plus variants/clones) using both SAD1024A and MN3007/MN3207 and the Reticon versions generally sound somewhat clearer/more open in the highs, perhaps more "transparent" or "airy".  However, any such benefit would be mostly limited to flangers (or extremely short delays), as the heavier filtering required by longer delay times would tend to negate any high frequency response advantage.  Anyone seeking a grungy, low-fi, sounding long delay would likely see little benefit.  This could easily be tested by someone with a breadboard and simple short delay circuits for the SAD1024A and MN3007 and running a frequency response sweep, noting the difference in drop-off as frequency increases. 

[ElectricDruid]

Does anyone know enough about the construction of these BBD chips to say *why* the Reticon ones have such a different clock pin capacitance than the Panasonic/Matsushita/Shanghai Belling ones?

For bonus points, why did the modern clones copy the ones with *higher* clock pin capacitance rather than lower?

If someone came up with an MN3207 clone with a new, modern, super-low clock pin capacitance, they'd have a winning chip on their hands. Santa, if you're listening...

[Eb7+9]

I may have solved this problem a few months ago ...

using clock pin capacitance of 10nF (outragiously high) I was able to preserve a nice crisp
clock wave form well above 4Mhz using a new approach

here's a reproduction of SANGSTER's work on the simulator:



my advice would be not to blame the technology ... it's really a simple engineering problem

---

btw, are there any references to this parallel/multiplexing business ??!

[ElectricDruid]

I may have solved this problem a few months ago ...

using clock pin capacitance of 10nF (outragiously high) I was able to preserve a nice crisp
clock wave form well above 4Mhz using a new approach

here's a reproduction of SANGSTER's work on the simulator:

Sorry, you have to tell me a bit more about what I'm looking at/for, 'cos otherwise it's just a pretty diagram. I see something that looks like a sim of a BBD, and some outputs that look like a very high frequency clock, and a sine input and a bent sine output. So far, so normal. What's special? Because it doesn't leap out.

Was there a thread about this I missed? Thanks!

my advice would be not to blame the technology ... it's really a simple engineering problem

Yes, I likely agree!

btw, are there any references to this parallel/multiplexing business ??!

Some of the SAD chip datasheets show applications using the two BBDs on a chip in parallel. The idea was that differential signals would cancel some of the noise and give better S/N.

[Mark Hammer]

Does anyone know enough about the construction of these BBD chips to say *why* the Reticon ones have such a different clock pin capacitance than the Panasonic/Matsushita/Shanghai Belling ones?

For bonus points, why did the modern clones copy the ones with *higher* clock pin capacitance rather than lower?

If someone came up with an MN3207 clone with a new, modern, super-low clock pin capacitance, they'd have a winning chip on their hands. Santa, if you're listening...

They would, but then I suppose they wouldn't have a 3207 clone.  Perhaps other folks have some insight into the protocols of leasing a die for a chip.  I know nothing about the I.P. law around such things, and what is required to call something a clone of a 3207, as opposed to simply a pin-for-pin functional replacement.  I am told that the Beling and Coolaudio 3207s are a little different, though I have no idea why.

[Eb7+9]

I know from experience that a SAD1024 running at 500kHz or higher in parallel-multiplex mode with relaxed filtering easily reproduces beyond 10kHz, as cymbals are surprisingly well resolved in the isolated delay path (the overall BBD path fidelity surprised me through AKG K702 headphones... a bit noisy, but impressively clear). 

makes sense now - from the 1024 datasheet I can see that parallel-multiplex averages the output of successive half-cycle samples, each side with its own built-in noise error voltage ... in principle this technique averages the sum of noise voltages also, which leaves that noise contribution the same as if we were going single-channel ... but the averaging of successive samples in itself forms a noise error relative to the true half-point values of the signal // ie/. not really sampling at twice the frequency but rather inferring a value ... this noise voltage is then added onto the noise voltage you get in non-duplex mode ... so, if I get this right we end up doubling bandwidth (thus relaxing nyquist filter requirements) at the expense of increase in sampling related noise - getting worse at higher frequencies

Anyone seeking a grungy, low-fi, sounding long delay would likely see little benefit.  This could easily be tested by someone with a breadboard and simple short delay circuits for the SAD1024A and MN3007 and running a frequency response sweep, noting the difference in drop-off as frequency increases. 

... is there a legit reason why these won't sample up to Fc/2 (at 1Mhz) without aliasing ?!

[ElectricDruid]

... is there a legit reason why these won't sample up to Fc/2 (at 1Mhz) without aliasing ?!

Not "aliasing" exactly, but if the clock pulses aren't clean and sharp, there's certainly distortion, and it probably affects higher frequencies worse than lower ones (bigger changes from one sample to another will be hit worse - and those are high frequency edges). So as Mark said, the clock drive matters *a lot*, especially on chips with more clock capacitance. Th thing about the SAD's low clock capacitance was mostly just that it made a lot of this stuff a lot easier, so people never had a problem they needed to solve.

If someone came up with an MN3207 clone with a new, modern, super-low clock pin capacitance, they'd have a winning chip on their hands. Santa, if you're listening...

They would, but then I suppose they wouldn't have a 3207 clone.  Perhaps other folks have some insight into the protocols of leasing a die for a chip.  I know nothing about the I.P. law around such things, and what is required to call something a clone of a 3207, as opposed to simply a pin-for-pin functional replacement.

I very much doubt there's any IP reason, given the timescales involved. The SAD design (for example) must be 40+ years old. The reason there's been a sudden boom in analogue synthesizers and analogue synth chips just recently has a lot to do with patents and copyrights running out on most of those original designs - and they date from the 1980's. BBD's are 1970's tech, so they must be pretty much in the clear these days. Additionally, the internal circuit must be so damn simple that I'd be surprised if anyone who works on things like this needs much of a clue. There's a diagram in the datasheet which I suspect is pretty much exactly what's on the silicon - it's just repeated a lot of times.
A "functional clone" is the only kind I think I care about - "Can I plug it in and it does the same job?". If I can plug it in and get an improvement in S/N ratio, or a wider bandwidth, or a easier way to push the chip to super-short flanger ranges, I'm all for it. Over-priced original chips can then be used to keep vintage devices in original condition and working order.

Basically, we're still building tons of stuff based on the MN3207, choruses, flangers, whatever - there's a ton of it out there. And mostly, those designs are now using modern-production chips. So why don't we have any modern production chips that out perform the specs from the 1970s? Is is really that hard to do? Are there technical reasons why modern processes aren't so suitable for this sort of stuff? (Does smaller geometry make it harder?)

I'm just curious why there are modern BBDs, but no *better* modern BBDs.

[Mark Hammer]

Most BBDs are perfectly acceptable for chorus, vibrato, and delay, as is, and do not require any additional buffering unless one intends to cascade a bunch of them to attain very long delays (e.g., four MN3205 units to get 1200msec of delay), such that a common clock is attempting to feed multiple sets of clock pins.  I suspect the reason why an "uber-3207" hasn't emerged is because the advantages of very low clock-pin capacitance really only occur in the context of flanging, and flangers are like the girlfriend's friend who tags along and has "a really nice personality", but no one wants to date her (some of us really like flangers, and the others say "I don't get it."). 

[R.G.]

The SAD1024 was intended to be able to delay video as well as audio. The clock capacitance and speeds were probably a side effect of the chip designers solving the video delay problems. It's loafing when sampling and delaying audio. The 3xxx chips were designed for delaying audio, so they were solving a different problem, probably more an economic problem than a performance one.

The limits and performance of chips, especially PMOS, NMOS and the !!! CMOS of the time were dictated by the chip fabrication of the time: how small can you make a MOSFET on a silicon wafer, how fast and low capacitance can the doping process produce, can you use poly-crystalline epitaxial gates, or must you use metal gates, what's the biggest capacitance (for storing a switched-capacitor voltage), how low an Rdson can you make for the transfer MOSFETs, and how much gate capacitance do these have, and on and on and on. The limits and advances of silicon semiconductor fabrication seem quaint these days, but they were a real and constantly shifting factor in what could be done as the chip designs marched on. The SAD1024 was remarkable for what it achieved - for the time.

SAD1024s were not very reliable. They sometimes just die for no obvious reason. I've always put this down to them being at the edge of what was possible in semiconductor fabs when they were made. I once found some SAD1024s in a surplus outfit, recognized them for what they were, and bought their entire stock, some several hundred chips. I wound up retailing and wholesaling them away over the years. Many of these went to pedals which had simply stopped working because the 1024 inside died. The virgin tubes of SAD1024s were about a 30% non-functional. I had to test them every one, excepting to the buyers who bought large quantities and haggled a lower price untested.

Reticon once made an SAD4096, a quad-length variant of the same technology. The only ones I've ever heard sound horrible. The processes in the storage and chaining inside the cells suffer from clock leakage and "diffusion", a seepage of charge from adjacent cells as the charges are passed down the line. Apparently this got severe in the 4096, which is another way of saying that the semiconductor processing was not advanced enough.

To answer a few questions -
Why don't we see much-improved SAD1024s today? The semiconductor processes are vastly more capable. But the pure digital alternatives have gotten even better than an analog bucket brigade could hope to be, and more importantly to semiconductor makers, a far higher yield from each wafer, as they don't rely on the analog qualities of the wafers and diffusion. Sure, SAD1024s could be redesigned into new process lines, and manufactured. PAYING for the design into the new fab processes with the tiny (to semiconductor makers) market means that this is not going to happen. It's most likely not a technical limitation, but an economic one. Pedal makers simply don't have enough volume to get their attention. Shoot, the PT2395 and FV-1 and families are bit players.

And as to a modern bucket brigade... the FV-1 will do any of that you want, and do it in stereo as well as lots of other stuff. Mostly, you stick a crystal on an FV-1 and treat it like an analog part. Yeah, I know some of its limitations, but you can sure flange with it and do analog delays with bandwidth above any frequency a guitar speaker can reproduce.

... is there a legit reason why these won't sample up to Fc/2 (at 1Mhz) without aliasing ?!

Probably. I don't know exactly what you're chasing, but in the real world, not aliasing at Fs/2 is not going to be achieved in practice. Unless the signal is inherently band limited to under Fs/2 and preferably well under that, there will be some spillover above Fs/2. Analog filters are just not brick walls, even fancy Eliptical-Cauer-Whitsontide-St.Swithuns filters. In practice, never assume you'll be free of aliasing anything above Fs/3. It's the nature of the beast. As Yogi Berra said, in theory, theory and practice are the same. In practice, they're not.

[12Bass]

Most BBDs are perfectly acceptable for chorus, vibrato, and delay, as is, and do not require any additional buffering unless one intends to cascade a bunch of them to attain very long delays (e.g., four MN3205 units to get 1200msec of delay), such that a common clock is attempting to feed multiple sets of clock pins.  I suspect the reason why an "uber-3207" hasn't emerged is because the advantages of very low clock-pin capacitance really only occur in the context of flanging, and flangers are like the girlfriend's friend who tags along and has "a really nice personality", but no one wants to date her (some of us really like flangers, and the others say "I don't get it."). 

I've always loved flangers.  That swoosh over the top is to die for!  Such a dramatic, exciting sound.  Chorus is nice, but pretty bland in comparison.  Bought my SAD1024A from Radio Shack around 1980 and wanted to make a flanger with it, but the only easily available project was Radio Shack "Electronic Reverb", which was made for line level chorus and delay, with regeneration for metallic reverb; it wasn't clocked fast enough for flanging and the results were disappointing. 

It wasn't until around 2010 that I used that Reticon chip in an A/DA Flanger clone and got it sweeping up to 1MHz where it really shines.  Sad that it died after my friend's son knocked it over a few years ago.  My flanger dreams were dashed after over forty years.  I have a feeling the SAD1024A got damaged somehow but haven't had the heart to troubleshoot it. It gets pretty warm while in the circuit, but the delay path is silent.  Perhaps at some point you guys might help me get it going again? 

[amptramp]

The technology still exists with CCD camera chips.  I have looked, but not seen any camera chips with a test input that allows you to input an electrical signal and check the entire CCD delay process but if you have the patience and can get some large 1 megapixel camera chips from long ago (but later than most BBD chips), you could still have an optical input limited to one pixel and control the readout frequency to get a variable delay or move the optical input to change the delay on a fixed-frequency readout sweep.

Even if you don't limit the input to one pixel, the readout will be in the same order all the time so you just have to open a sample-and-hold at a variable clock count to get a variable delay.  You can choose just the 543,768th output pulse then the 543,694th pulse to get the speedup part and the 543,937th pulse to get the slowdown part of the waveform.  All you have to do for the LFO is generate a bunch of numbers for the sample count that are in sinewave or triangular order and vary the differences between them to get the depth control.

There aren't many BBD's still around for audio and not many sources of new ones.  But there are tons of old CCD camera chips and if you are not going into production, lots of cameras being sold cheap at garage sales because they are old 1 or 2 megapixel technology.  Maybe you can still get the SAD1024 sound with electronics from this millenium.

[Mark Hammer]

In fact, if one inspects the datasheets for the SAD1024, one of the suggested applications IS for video-processing, so I imagine it was designed with very low clock-pin capacitance to serve the clock frequencies needed for 1980's video processing.

All of that said, I listened and watched, slackjawed, as Mike "where oh where are you" Irwin demoed an MN3007 for me, being clocked up to 1.5mhz.