THAT vs TI

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
@ CMRR many of us in the industry are well familiar with the history of THAT corp, and before that DBX going back to Dave Blackmer's early VCA development. Dave was working at API at the time but reportedly they weren't interested (n) .
During all the rumblings about console automation (Allison, Olive, etc.) in the early seventies, I had the inspiration for Compumix (based on years of working with telemetry systems) while I was chief engineer at Quad-Eight. But the hardest part was finding a good VCA. I had developed a novel PWM/PAM unit - it converted input audio into 1 MHz PWM (using a single transistor and a TTL Schmitt trigger IC) and then used a pair of complementary MOSFET switches to modulate the PWM's amplitude in response to the DC control voltage. Then a simple low-pass filter recovered the audio. It worked really well, including a low noise floor, but it was hard to get the zero-control-voltage output to under about -80 dB because N and P-channel MOSFETs of the day were awful compared to what's available today. We made about a hundred of them (I still have one as a souvenir) that were used as sub-group faders on a few large custom boards back in the day. But the trimming required ruled them out for the Compumix system, so I was told to reverse-engineer a dBx 202 and we made our own version of the module until dBx became aware and Quad-Eight got the customary "cease and desist" letter from dBx's attorney. They were gracious enough to give us a favorable quantity price and we used the 202 in production Compumix systems. Compumix, an "add-on" mixdown automation system sold well - to A&M, Warner Brothers, Armin Steiner, Neil Young (who cut his "Time Fades Away" album directly from multi-track via Compumix), and many others. Compumix's "update" function apparently established a standard for using -15 dB as a reference point for updating a track's gain trajectory. Bud Bennett, owner of Quad-Eight back then, was notoriously stingy but, to my astonishment, gave me a bonus of half my yearly salary that year for developing such a successful product! It certainly wasn't trivial back in the day before floppy discs and single-chip processors!
 
Indeed VCA are obsolete when digital audio paths are being used, a simple digital multiply can vary gain up or down with no distortion or added noise.

We still live in a physical world so mic preamps will remain in use for some time. Of course the first company that combines a microphone with A/D convertor, with no signal integrity compromises will dominate, but that is still a wet dream.


hey JR

do you mean like this?

These ICs are designed for digital live consoles

http://thatcorp.com/626x_Dual_Programmable_Preamp_ICs.shtml
 
There is a big PRACTICAL difference between the THAT and TI driver chips that's not apparent from da datasheets.

It's to do with what happens if you feed an unbalanced input. Wayne of proaudiorecording (IIRC) did & posted extensive tests but I can't remember where. JohnRoberts may remember the exchange and the links. There's loadsa factors so ...

The executive summary is to use the THAT driver & input chips ... if you want sensible behaviour under ALL conditions.
 
During all the rumblings about console automation (Allison, Olive, etc.) in the early seventies, I had the inspiration for Compumix (based on years of working with telemetry systems) while I was chief engineer at Quad-Eight. But the hardest part was finding a good VCA. I had developed a novel PWM/PAM unit - it converted input audio into 1 MHz PWM (using a single transistor and a TTL Schmitt trigger IC) and then used a pair of complementary MOSFET switches to modulate the PWM's amplitude in response to the DC control voltage. Then a simple low-pass filter recovered the audio. It worked really well, including a low noise floor, but it was hard to get the zero-control-voltage output to under about -80 dB because N and P-channel MOSFETs of the day were awful compared to what's available today. We made about a hundred of them (I still have one as a souvenir) that were used as sub-group faders on a few large custom boards back in the day. But the trimming required ruled them out for the Compumix system, so I was told to reverse-engineer a dBx 202 and we made our own version of the module until dBx became aware and Quad-Eight got the customary "cease and desist" letter from dBx's attorney. They were gracious enough to give us a favorable quantity price and we used the 202 in production Compumix systems. Compumix, an "add-on" mixdown automation system sold well - to A&M, Warner Brothers, Armin Steiner, Neil Young (who cut his "Time Fades Away" album directly from multi-track via Compumix), and many others. Compumix's "update" function apparently established a standard for using -15 dB as a reference point for updating a track's gain trajectory. Bud Bennett, owner of Quad-Eight back then, was notoriously stingy but, to my astonishment, gave me a bonus of half my yearly salary that year for developing such a successful product! It certainly wasn't trivial back in the day before floppy discs and single-chip processors!
I killed my share of brain cells on my bench trying to make gain cells with wide dynamic range, low distortion, low noise, and good control laws, back in the 70s-80s. Most decent approaches tagged some but not all of those bases. I never tried to make analog gain control using PWM, but did see PWM used to control multi-pole filters by duty cycle modulating the Rs in multistage RC filters. I suspect it had issues like class D audio amps where noise in the comparator, and PS (control voltage?) talk directly into the audio output.

I gave up trying to roll my own back then and used Paul Buff's VCA back in the early 80s. At Peavey I even patented an unremarkable class A (noisy) VCA whose claim to fame was using unselected parts with no trimpots. Factory trims at Peavey were discouraged. ;) Later when the industry was free to purchase Dbx/THAT VCAs, Peavey did too.

Welcome to the forum, I look forward to hearing more of your industry experiences.

JR

PS: I was an old analog dog who didn't start coding microprocessors until this century.
 
PS: I was an old analog dog who didn't start coding microprocessors until this century.
It's good to meet you John, if only virtually. I'm still "an old analog dog" whose found enough interesting consulting work (even today) to stay completely focused on analog. I guess I'm "old school" in a lot of ways - being dragged, kicking and screaming, into the 21st century. But, as I see it, there's still lots of room for innovation in the analog world - "digital" doesn't necessarily mean better. Although I once wrote 50 lines of code for a 8051 back in the 80s, I've never had an interest in writing code.

Incidentally, I became fascinated with PWM after first learning about it in 1965 - and later designed a 30-watt DC-coupled "audio" amplifier switching at 100 kHz in 1974 using the first commercially available (I think) power MOSFET, the Siliconix VMP-1. I was working for the laser-light show company Laserium at the time and we used hundreds of them to drive the optical scanners in the laser projectors. Todays semiconductors make PWM designs even better. I'd love to revisit my old PWM/PDM VCA design using semiconductors 50 years newer!
 
There is a big PRACTICAL difference between the THAT and TI driver chips that's not apparent from da datasheets.

It's to do with what happens if you feed an unbalanced input. Wayne of proaudiorecording (IIRC) did & posted extensive tests but I can't remember where. JohnRoberts may remember the exchange and the links. There's loadsa factors so ...

The executive summary is to use the THAT driver & input chips ... if you want sensible behaviour under ALL conditions.
Thanks Ricardo! If you'd like to see 'scope trace proof of misbehavior in the SSM2142 and DRV134 parts, check the waveforms at OutSmarts Line Driver Clipping Behavior. Yes, that's ultrasonic oscillation on the negative swing of the SSM2142!
THAT did a wonderful thing with their patented solution to this problem with the original SSM2142 (and as copied by Burr-Brown/AD) ... the THAT drivers deliver what the other guys promised ...
 
It's good to meet you John, if only virtually.
same here
I'm still "an old analog dog" whose found enough interesting consulting work (even today) to stay completely focused on analog.
and I've watched analog console companies get creamed by digital bang for the buck.
I guess I'm "old school" in a lot of ways - being dragged, kicking and screaming, into the 21st century. But, as I see it, there's still lots of room for innovation in the analog world - "digital" doesn't necessarily mean better. Although I once wrote 50 lines of code for a 8051 back in the 80s, I've never had an interest in writing code.
The world is analog, but digital gets close enough to perfect for government work.

I didn't start designing with microprocessors because they were digital, I did it because the ability of digital circuitry to be programed to use conditional decision making is incredibly powerful. I started out trying to design my drum tuner using analog circuitry and it soon became unwieldy. A simple 8 bit micro could do things cheaper and so much better than any analog circuit I could imagine (even though analog was my preference).
Incidentally, I became fascinated with PWM after first learning about it in 1965 - and later designed a 30-watt DC-coupled "audio" amplifier switching at 100 kHz in 1974 using the first commercially available (I think) power MOSFET, the Siliconix VMP-1.
I made a crude very low power class D amp on my bench in the early 70s, just to see if I could. Not much power, but it was picked up on every radio in the building. :unsure:

Decades later I had the bad luck to be product manager over Peavey's class D amplifier program, that was clearly decades before it's time, or decades before the switching device technology's time... My life would have been so much easier if we had modern switching devices back then. Speaking of class D what do you think of Bruno Putzey's class D approach? (I like it).

I was working for the laser-light show company Laserium at the time and we used hundreds of them to drive the optical scanners in the laser projectors. Todays semiconductors make PWM designs even better. I'd love to revisit my old PWM/PDM VCA design using semiconductors 50 years newer!

Exotic, esoteric analog audio is a limited and probably still shrinking market. The best buggy whip ever is still a buggy whip... but that could be fun. I have some old analog circuits that I know I could do better but I don't think the market cares.

JR
 
Last edited:
To summarise the recent discussion, would it be true to say that there are known issues with the TI OUTPUT driver devices but the iINPUT devices are OK?

Cheers

Ian
 
Indeed VCA are obsolete when digital audio paths are being used, a simple digital multiply can vary gain up or down with no distortion or added noise.
There _is_ still a distortion product added by the multiply in many systems. For example: multiply two 32 digit numbers. You get (potentially) a 64 digit number. Most practical systems will go ahead and round them back down to 32 digits. You'd have to quiz the compiler guys on the particulars, but that rounding _is_ distortion. If you do the programming right it stays small, and hey, at least it's not polluting the signal with _externally_ generated noise. :)
Of course the first company that combines a microphone with A/D convertor, with no signal integrity compromises will dominate, but that is still a wet dream.
'No signal integrity compromises' could be hard to apply to even the best VCAs though. Usually a pot does way better. Plenty of digital out mics out already too. There's an AES spec.

In theory though, let's do a quick calculation of the dynamic range required...eh?

A 150Ω source at 275°K is about -133dBV.

I happen to know the maximum an SM58 can put out (you need a very loud source, naturally) is about a volt. 0dBV

58s are perhaps a *bit* higher output impedance, but whatever. You maybe get about 130dB of dynamic range from a typical 150Ω dynamic mic. The latest generation of converters are basically there now.
 
You'd have to quiz the compiler guys on the particulars, but that rounding _is_ distortion. If you do the programming right it stays smal

You don't have to be a compiler guy to figure out that it's one bit out of 64 which is infinitesimally small to the point of being completely insignificant. Digital is very good at amplification, attenuation, summing, EQ, delay and things that are simply mathematical transformations. No VCA or analog circuit will ever beat digital at those things.

However, sometimes there are subtle imperfections in circuits that people find desirable like tube distortion and the wavering oscillator of a CS-80 and so on. For those things digital is a crude approximation of reality and can easily reach processing limits. GPU programming is probably the next big thing in digital audio processing but you still need algorithms that accurately model those imperfections and I don't suspect there are many, if any, people in this world that understand such things and happen to write efficient parallel processing code.

It is borderline trivial to put an ADC in a mic and get pristine audio. At least the electrical aspects are not difficult. The problem is that there is potentially a failure from the ADC delay and a probably more serious problem is the negative sales mojo of a USB cable sticking out of a mic.

You maybe get about 130dB of dynamic range from a typical 150Ω dynamic mic. The latest generation of converters are basically there now.

The noise floor of a mic in anechoic chamber is going to be well above that of a mediocre ADC.
 
Sorry about the veer...

A digital multiply returns more resolution than the input terms. There is no way to characterize that as distortion. It is pretty much the opposite of distortion (more accuracy), while it can be hard to use all that extra resolution effectively.

JR
 
Resolution and accuracy are not the same thing.

And this increased resolution does need to be reduced to a useable word length, which results in distortion.

Good designers will minimize this through judicious use of dithering.
 
Resolution and accuracy are not the same thing.
Agreed
And this increased resolution does need to be reduced to a useable word length, which results in distortion.
Even if you just truncate the entire extra resolution that only amounts to less than one least significant bit. Any typical source program will have a noise floor well above that LSB.
Good designers will minimize this through judicious use of dithering.
Dither a 32x32 bit multiply? The noise floor of the program will dither that wether you want it to or not.

Again I apologize for this veer.

JR
 
The executive summary is to use the THAT driver & input chips ... if you want sensible behaviour under ALL conditions.

If you'd like to see 'scope trace proof of misbehavior in the SSM2142 and DRV134 parts, check the waveforms at OutSmarts Line Driver Clipping Behavior. Yes, that's ultrasonic oscillation on the negative swing of the SSM2142!
Yes. That's the main fault but IIRC, Wayne found some other stuff too .. especially on long lines.
To summarise the recent discussion, would it be true to say that there are known issues with the TI OUTPUT driver devices but the iINPUT devices are OK?
Ian, if you are going to sully your vacuum path with EVIL silicon, why not use the best. ie THAT OutSmarts & Guru Whitlock's InGenius hand carved from solid BS by Milford virgins? :)

... unable to resist the veer ..

I did the work that led to Calrec adopting the Blue (Allison, later Valley People IIRC) VCAs instead of the Black (DBX) ones. They were used in most big Calrec desks of the early 80s cos they had advantages feeding the virtual earth mixer on large desks and also on the 'Assignable'; a 72 channel computer controlled desk with a small (what I think would be called today a) 'control surface'. There were 2 big fridge size racks in the next room with the actual gubbins. Da intelligence was provided by a 6809 development kit with 8" floppies :eek:

You need to dither if you do any non-trivial DSP operation on an integer DSP system. A fade counts as non-trivial .. as does each butterfly on an in-place FFT. Today, you would do everything in floating point and only dither when you get out of the box .. when you save the file or play it.

But don't underestimate the difficulty of doing proper dithering and/or anti-aliasing. There's a page somewhere which shows the artifacts from various Sample Rate Converters. Some very big names result in crud which my DBLT panel can pick out on 16b paths.

... old geezer who has done time for DSP on early integer DSP chips :)
 
Last edited:
ts it surrounded by siliconIan, if you are going to sully your vacuum path with EVIL silicon, why not use the best. ie THAT OutSmarts & Guru Whitlock's InGenius hand carved from solid BS by Milford virgins? :)
It is not my signal path in this case. I am designing an EQ for a customer who wants it surrounded by silicon balancing and de-balancing. I am going for a -6dB input device because the poor thing does not have the headroom you get with a 250V HT rail. I thought if I have to from here first.visit the dark side I might as well get the best advice from here first.

Cheers

Ian:cool:
 
Even if you just truncate the entire extra resolution that only amounts to less than one least significant bit. Any typical source program will have a noise floor well above that LSB.

That is a bit of a misconception, noise floor in and of itself is not always sufficient. It needs to be added as an uncorrelated source at the time of bit reduction.
 
BUT we are (somebody is) talking about a 32b x 32b multiply.... I can't imagine a path quiet enough to not dither that, and/or hearing audible artifacts down 32 bits.

My understanding about modern oversampling A/D convertors is that they are dithered by their own analog front end stage's noise floor.

I am old enough to remember when dithering was useful and used to chase away quantization distortion in lower bit system paths.

You may be correct in theory but in practice I do not perceive the big problem.

JR

PS: When dealing with large analog summing systems we deal with much larger known nonlinearities and deviations from ideal. Ironically perhaps I have an improved analog summing system approach (current source summing), that I abandoned as moot because of the near ideal summing in the digital domain.
 
With due respect to all participants in THAT vs. SSM/DRV discussion , I would like to state that

- SSM and DRV datasheets (especially) are far more extensive and detailed than with THAT IC,
- THAT paper did not write at which supply voltage critical oscillograms were obtained
(THAT output voltage is limited to 16V, on SSM and DRV on 17.5V ??),
- and it is also far more important to me personally how these ICs behave in normal operating conditions rather than in a situation that so rarely happens and which is the result of an audio engineer’s mistake. By that I mean overload and clipping, not single-ended mode of operation. In these situations, the only thing that matters is that the IC survives such misuse.

I have used SSM ICs many times and THAT several times in my projects and I have not noticed any difference in normal operation.

P.S. It is rather strange that in the THAT paper they said in a footnote
"** Not specified by TI or ADI, figures based on THAT measurements", and complete diagrams are available to them. THAT did not display any THD vs. Freq diagram, BTW.
 
- SSM and DRV datasheets (especially) are far more extensive and detailed than with THAT IC,
If extensive and detailed data sheets convince you to prefer a product, I hope you use Jensen transformers in all your designs! ;) I know of no other maker that comes close! I created them and was always frustrated that my competitors conveniently over-simplified their data sheets - most often omitting important low-frequency level-handling data or failing to disclose source impedance for distortion data. Sorry, but it's a pet peeve (by the way, I no longer have any financial ties to Jensen since selling it 7 years ago).

I'd also like to point out that TI does some truly awful data sheets now and then. I have a couple of clients that use the TLV320 series CODEC. The data sheets go on and on for over 100 pages about the digital specs, but the analog specs are horribly sketchy and incomplete. And, if you ask TI for some details, they just say "sorry, those parts were designed by engineers that don't work here any more." Another pet peeve!
 

Latest posts

Back
Top