Researching state of the art audio ADC performance

[AnalogSteph]

I've been digging through ADCs old and new lately, which had me wondering:
What's the real SOTA in audio ADCs?

Nowadays, if you want to buy a fancy ADC, you might choose e.g. an RME ADI-2 Pro FS R Black Edition (quite the mouthful), which sports these ADC specs:

Input sensitivity switchable +24 dBu, +19 dBu, +13 dBu, +4 dBu @ 0 dBFS
Signal to Noise ratio (SNR) @ +13/19/24 dBu: 120 dB RMS unweighted, 124 dBA
Signal to Noise ratio (SNR) @ +4 dBu: 119 dB RMS unweighted, 123 dBA
THD @ -1 dBFS: -116 dB, 0.00016 %
THD @ -10 dBFS: -125 dB, 0.000056 %

Which is absolutely first-rate, of course - I think I've seen one other ADC with a -117 dB THD spec, but that's about it. Some ADCs have a rated dynamic of up to 130 dB(A) these days, but the RME is all but inflexible with its variable input sensitivity and allows covering a total range of 139 dB / 143 dB(A). Trying to get 120 dB or more through a studio seems like a bit of a challenge anyway.

Now what if I told you that there is a DAC released in 1998 with the following specs among others:

Dynamic range or signal-to-noise ratio (input sensitivity:+28dBu=0dBFS, 1kHz@-60dBFS):
typical: 131.5dB (unweighted RMS)
worst case:
>129.0dB CCIR-RMS
>131.0dB (A weighted RMS)
>129.0dB (unweighted RMS)

Dynamic range or signal-to-noise ratio (input sensitivity:+18dBu=0dBFS, 1kHz@-60dBFS):
typical: 128.0dB (unweighted RMS)
>127.0dB CCIR-RMS
>129.0dB (A weighted RMS)
>127.0dB (unweighted RMS)

Total harmonic distortion and noise (1kHz@ -1dBFS):
typical, <-108.0dBFS (0.00045%) (unweighted RMS)
worst case,<-105.0dBFS (0.00063%) (unweighted RMS)

Intermodulation distortion:<-90dB
Note: The intermodulation test from AES17 uses 18kHz and 20kHz signals at -6.03dBFS each. The result is the ratio of the total output rms signal level to the rms sum of the 2nd and 3rd order modulation products at 2kHz and 16kHz.

Spurious aharmonic levels:<-130dBFS(1kHz@ -1dBFS)
Note: Highest level of any aharmonic spurious component

Any spurious levels:
<-112dBFS signals to -1dBFS
<-130dBFS signals below -20dB FS
<-140dBFS signals below -60dB FS
Note: Highest level of any harmonic distortion component

Clearly the distortion department has still made some progress since then, although even those numbers would still be very good, and at -60 dBFS a good modern converter would probably show basically no harmonics at all. It's also 24/96 "only" (the latest and greatest back then). But an instantaneous dynamic range of over 130 dB unweighted is nothing short of remarkable. That's DC-to-daylight level. Clearly no typical monolithic CMOS ADCs in this one (about the best one you could get back then was the Crystal CS5396, 117 dB unweighted with a following wind).

What it is? Meet the Prism Sound Dream AD-2. Not sure what these cost back then but you can buy one for something like 9 grand now. I would certainly love to interview the designers; they clearly were very proud of their product (and deservedly so).

So can you do better these days? After all, IC-based ADCs have caught up and several are sporting up to 130 dB as well these days. Possibly. The (slim) specs for a Weiss ADC2 indicate the following at +26 dBu in:

THD+N at 1 kHz:
Less than −103 dBFS at −3 dBFS output level, unweighted

SNR at −40 dBFS input:
Higher than 110 dB unweighted

So that might indicate a dynamic range of about 150 dB, unless I am misinterpreting things. (Which would be super high but definitely not impossible at +26 dBu, though I suppose internal levels may be lower than that.) Makes you wonder whether you can actually get converters this good these days or whether it's some sort of composite setup.

Speaking of which, composite ADCs seem to be dating back at least as far as 1994:
Studio Sound, March 1995

DG 4D goes to third generation
The Deutsche Grammophon Recording Centre has developed a third generation upgrade of the Stage Box system central to the 4D recording chain. All recordings made by the Recording Centre since October 1994 have used the new DG AD III technology, whose convertors feature the new Crystal CS5390 delta-sigma 20-bit A-D convertor ICs to provide 23-bit digital-floating delta-sigma A-D conversion. The process employs two 20-bit convertors, one handling the input signal at unity gain and the other operated with 18dB gain. A sophisticated DSP algorithm regulates the crossfade between the two convertors, producing three bits of supplementary resolution. The DSP program was modified to allow the DSP chip to handle 20-bit convertors at its inputs and a 24-bit wordlength at its outputs. Quoted specifications include THD+n of -121dBFs with an input of 997Hz at -30dBFs and linearity errors within 1dB down to -135dBFs, together with a largely flat noise-spectrum. A further improvement is the development of the Authentic Clock Recovery system, permitting superior reconstruction of the master clock signal under real world operating conditions such as long cable runs and numerous interconnected PLLs, where phase modulation of the clock, jitter, becomes a limiting factor on overall system performance. Because Authentic Clock Recovery uses crystal PLLs driven at 512Fs, as opposed to the current 256Fs standard, A-D conversion at up to 96kHz is possible, with full oversampling capability.
Deutsche Grammophon, Germany. Tel: +49 4044 181115.

The CS5390, by itself, was a 48 kHz 20-bit delta sigma ADC rated 110 dB(A) DR and -100 dB THD+N and released in October of 1993. I didn't think you could just clock one twice as fast and expect it to work, but apparently it did (I suppose some debugging in the lab may have been quite necessary).

On a side note, I found out that even kilobuck ADCs do not necessarily sport a decent, halfway comprehensive set of specs, especially these days. The Mytek Brooklyn ADC's manual has got to be one of the worst offenders, it mentions a 130 dB dynamic range and <1 ps clock jitter spec but no mention of levels at all except for a mysterious "headroom: 13-20" (does that mean a +17 dBu to +24 dBu 0 dBFS level?). Relevant information is also strewn across the whole thing. This is in stark contrast to, say, RME manuals, or what you see in the '90s.

 

[RayDunzl]

What's the real SOTA in audio ADCs?

60~70dB once you actually record something from the air, I suppose.

 

[Blumlein 88]

60~70dB once you actually record something from the air, I suppose.

Which is why there isn't a big push to plumb the physical limits of ADCs.
Don't know I've ever seen any specs, but along with the Prism Dream, many consider the Pacific Microsonics to be the best of the best ever. I do seem to recall a loopback with a PM model 1 ADC and DAC hit something like -120 db nulls with music. I don't recall anything else getting close to that that I've ever heard about or read about.

 

[RayDunzl]

How does the Audio Precision accomplish its analog to digital conversions?

That would have to be SOTA+ in order to be able to measure SOTA devices...

 

[Blumlein 88]

http://danmon.dotnet.dir.dk/NTP/Mar...ures_and_leaflets/AX32_DX32_brochure_2018.pdf

These are considered top converters for lots of reasons well beyond the basic ADC/DAC in them. But that ADC/DAC is said to be 128 db Dynamic Range and THD below -120 db. Like many converters very incomplete and poorly specified conditions.

 

[Blumlein 88]

https://www.akm.com/content/dam/doc.../audio-adc/ak5578en/ak5578en-en-datasheet.pdf

https://www.akm.com/global/en/products/audio/audio-adc/ak5578en/

8 channel AK5778en which is the best AKM adc at the moment.

-112 db thd, 121 db SNR (you'd expect dynamic range to equal to slightly exceed this). This is an 8 ADC chip so you can parallel them for up to an SNR/Dymamic range of 130 db.

The Antelope Amari uses the AKM5778en. One chip per channel claiming an achieved 128 db dynamic range.
BTW Antelope is using CS43198 DAC chips (4 per channel) they claim 138 db dynamic range for playback.
https://en.antelopeaudio.com/products/amari/

Merging uses this same chip in the Horus and Hapi claiming similar specs.

Couple these with say an Earthworks ZDT microphone preamp and the ADC isn't missing much.

 

[AnalogSteph]

http://danmon.dotnet.dir.dk/NTP/Mar...ures_and_leaflets/AX32_DX32_brochure_2018.pdf

These are considered top converters for lots of reasons well beyond the basic ADC/DAC in them. But that ADC/DAC is said to be 128 db Dynamic Range and THD below -120 db. Like many converters very incomplete and poorly specified conditions.

Actually, this one is not too bad at all - the brochure alone has this to say:

• Dynamic range, ADC (A) > 123 dB
• THD + N, ADC (A) < -120 dB @ -6 dBFS (note level! probably higher at -1 dBFS)
• Dynamic Range, DAC (A) > 128 dB
• THD + N, DAC (A) < -115 dB @ -6 dBFS
• Cross talk (A) < -120 dB
• Mic Equivalent noise (A) < -130 dB @ 100Ω (note: equivalent to -129 dBu @ 150 ohm)
• Mic gain range - 18 to +70 dB
• Mic gain step accuracy < 0.25 dB
• Sample-rates 44.1 to 384kHz, DSD64, DSD128 (DoP)
• Clock accuracy < 2PPM via Digital PPL (that's neat! they must be using a really fancy master clock, too!)

And the manual adds these of interest:

THD+N(A): < -117 dB@-3dB FS / 0,00014%
Input Impedance (differential): > 10 k Ohm
Max input level: Adjustable from 9 dBu to 30 dBu in steps of 0.1 dB
Microphone input gain range/accuracy: Adjustable from -18 to +70 dB, in steps of 0.1 dB, ± 0.25 dB accuracy

Overall, well in the green despite some "number games".

I am fully aware that you probably don't need 130 dB of dynamic range in any non-pathological recording application. Directly capturing the output of a high dynamic range microphone under basically all conditions would have to be the main motivation indeed, and I presume that's why DG made their fancy composite ADC, too.

Don't know I've ever seen any specs, but along with the Prism Dream, many consider the Pacific Microsonics to be the best of the best ever. I do seem to recall a loopback with a PM model 1 ADC and DAC hit something like -120 db nulls with music. I don't recall anything else getting close to that that I've ever heard about or read about.

That would in fact be remarkable, since the Prism DA-2 only hit a dynamic range of a measly 111 dB unweighted - for some reason, DACs were substantially worse than ADCs at the time, in the IC world as well. You could definitely do it when combining the AD-2 with a substantially more modern DAC.

I've read about the Pacific Microsonics converters as well, but was never aware that they were this high-end.

 

[Wombat]

Year 2002.

 

[Blumlein 88]

To my knowledge, and I'm not that informed of microphone preamps the EW ZDT series might measure among the best.

https://earthworksaudio.com/products/preamps/zdt/zdt-1021/

Some games played there too. -120 db distortion is claimed. The EIN is obviously for something stupid like 1 ohm or a dead short.

 

[AnalogSteph]

To my knowledge, and I'm not that informed of microphone preamps the EW ZDT series might measure among the best.

https://earthworksaudio.com/products/preamps/zdt/zdt-1021/

Some games played there too. -120 db distortion is claimed. The EIN is obviously for something stupid like 1 ohm or a dead short.

Yes, that's shorted. Same as input noise density + A-weighting.

Very, very good either way. Given that output is up to +30 dBu and the .86 nV/√(Hz) at 20 dB gain translates to an EIN of -136.1 dBu (shorted), we may be looking at a 146 dB dynamic range there, and at unity gain, 160 dB (!). That's nuts. Even at +60 dB this one is still capable of 109 dB, or 100 dB with a nominal 150 ohm source, or 104 dB with a 50 ohm dynamic.

For comparison, the Millennia HV-3C that @Rja4000 has been measuring is right around -120 dBu EIN at 20 dB, for about 130 dB of dynamic range. And that's not exactly a slouch to begin with.

 

[pozz]

To my knowledge, and I'm not that informed of microphone preamps the EW ZDT series might measure among the best.

https://earthworksaudio.com/products/preamps/zdt/zdt-1021/

Some games played there too. -120 db distortion is claimed. The EIN is obviously for something stupid like 1 ohm or a dead short.

I've been looking at getting this one. Not sure how many other mic pres there are out hitting those kinds of numbers. Too bad the old Benchmark MPA-1 and PRE420 were discontinued.

 

[AnalogSteph]

Too bad the old Benchmark MPA-1 and PRE420 were discontinued.

While I'm sure the MPA1 would do well in tough environments with its CMRR trimming and all, going by EIN vs. gain it would be noticeably short even of the HVA-3C in terms of maximum dynamic range - slight extrapolation yields EIN of about -114 dBu at +20 dB, so ~121.5 dB. Still very good, but well short of the (admittedly insane) ZDT amps.

 

[pozz]

While I'm sure the MPA1 would do well in tough environments with its CMRR trimming and all, going by EIN vs. gain it would be noticeably short even of the HVA-3C in terms of maximum dynamic range - slight extrapolation yields EIN of about -114 dBu at +20 dB, so ~121.5 dB. Still very good, but well short of the (admittedly insane) ZDT amps.

I've been considering buying the ZDT just to send it for review, and either keep or sell depending on how well it does. The price tag is daunting though.

I also asked Benchmark if they were going to reintroduce mic pres. "Not this year," was the response. Somewhat encouraging.

 

[Blumlein 88]

I'm not sure I understand how you are getting your numbers for some aspects. @AnalogSteph

For instance, loading my Antelope mic pres with 100 ohms, the EIN is 2 db less than the noise for the resistor at any gain from 35 db up to 65 db. Below 35 db gain the device noise floor intrudes more. At 0 gain it is about 108 db. Basically appears you have a 108 db wide range which slides up and down with the gain setting. Max input on this pre is +9 dbu.

 

[AnalogSteph]

I've been considering buying the ZDT just to send it for review, and either keep or sell depending on how well it does. The price tag is daunting though.

Sure is. Not that much more than some other options with substantially smaller dynamic range though.

BTW, mic self-noise for the M30 would appear to be about -101 dBu(A) if I have my math right. A lot of that is likely to have a 1/f characteristic, so some experiments may be necessary to determine which level of amplifier white noise starts to degrade it. I almost can't imagine you need better than somewhere between -110 and -120 dBu EIN though.

What kind of maximum level are you going for? If you do need the rated 140 dB SPL, that's +17 dBu - not very much room for any gain there. Not a job for just any preamp if you still need dynamic range all the way down to self-noise. -110 dBu EIN at +7 dB gain and +17 dBu input level means 127 dB of dynamic range required (relative to +24 dBu output - so 133 dB relative to +30 dBu). Aside from getting you into trouble with ADI-2 Pro dynamic range limits.

I sort of doubt you need both ends of the mic's dynamic range at the same time though. Let's say we're talking 0.5-1 m measurements at up to 110 dB SPL or so, and suddenly things are a great deal more relaxed. ~100 dB at 37 dB of gain sounds a great deal more manageable.

 

[Blumlein 88]

Sure is. Not that much more than some other options with substantially smaller dynamic range though.

BTW, mic self-noise for the E30 would appear to be about -101 dBu(A) if I have my math right. A lot of that is likely to have a 1/f characteristic, so some experiments may be necessary to determine which level of amplifier white noise starts to degrade it. I almost can't imagine you need better than somewhere between -110 and -120 dBu EIN though.

What kind of maximum level are you going for? If you do need the rated 140 dB SPL, that's +17 dBu - not very much room for any gain there. Not a job for just any preamp if you still need dynamic range all the way down to self-noise. -110 dBu EIN at +7 dB gain and +17 dBu input level means 127 dB of dynamic range required (relative to +24 dBu output - so 133 dB relative to +30 dBu). Aside from getting you into trouble with ADI-2 Pro dynamic range limits.

I sort of doubt you need both ends of the mic's dynamic range at the same time though. Let's say we're talking 0.5-1 m measurements at up to 110 dB SPL or so, and suddenly things are a great deal more relaxed. ~100 dB at 37 dB of gain sounds a great deal more manageable.

I'm guessing Pozz is thinking of his Earthworks M30 with their preamp.

 

[pozz]

@Blumlein 88's got it.

 

[AnalogSteph]

I'm not sure I understand how you are getting your numbers for some aspects. @AnalogSteph

For instance, loading my Antelope mic pres with 100 ohms, the EIN is 2 db less than the noise for the resistor at any gain from 35 db up to 65 db. Below 35 db gain the device noise floor intrudes more. At 0 gain it is about 108 db. Basically appears you have a 108 db wide range which slides up and down with the gain setting. Max input on this pre is +9 dbu.

Yes, this fixed range kind of behavior is encountered where it's not input but rather output noise that is limiting dynamic range. A typical small mixer will have a tone control section following that generally is the bottleneck in terms of output noise floor, for example.

Depending on preamp construction, input noise itself may also degrade when turning down gain though, even before the maximum dynamic range "window" is making itself felt.

My calculations are simple:
By definition, EIN [dBu] = output noise [dBu] - gain [dB].
Also,
Output dynamic range [dB] = maximum output level [dBu] - output noise [dBu].
Hence,
Output dynamic range [dB] = maximum output level [dBu] - EIN [dBu] - gain [dB].
For the ZDT at 0 dB, that's +30 dBu - (-130 dBu) - 0 dB = 160 dB.
EDIT: For the ZDT at +20 dB, that's +30 dBu - (-130 dBu(A)) - 20 dB = 140 dB(A).
It's obviously (linearity!) the same as
Input dynamic range = maximum input level [dBu] - EIN [dBu]

I'm guessing Pozz is thinking of his Earthworks M30 with their preamp.

Corrected, not sure why I wrote E30.

 

[Blumlein 88]

Yes, this fixed range kind of behavior is encountered where it's not input but rather output noise that is limiting dynamic range. A typical small mixer will have a tone control section following that generally is the bottleneck in terms of output noise floor, for example.

Depending on preamp construction, input noise itself may also degrade when turning down gain though, even before the maximum dynamic range "window" is making itself felt.

My calculations are simple:
By definition, EIN [dBu] = output noise [dBu] - gain [dB].
Also,
Output dynamic range [dB] = maximum output level [dBu] - output noise [dBu].
Hence,
Output dynamic range [dB] = maximum output level [dBu] - EIN [dBu] - gain [dB].
For the ZDT at 0 dB, that's +30 dBu - (-130 dBu) - 0 dB = 160 dB.
It's obviously (linearity!) the same as
Input dynamic range = maximum input level [dBu] - EIN [dBu] snip...

The part I'm not sure is right is maximum input level minus EIN the way you are doing it. The measurement would indicate 30 dbu and 130 dbu lower for the EIN. A dynamic range at a given setting of 130 db. The 160 db would only seem the case considering the lowest level at low gain and the highest level at high gain. I believe you'll have a 130 db wide dynamic window however like my Antelope interface. I've not tested bunches of mic pres so maybe I have it wrong.

There is additional info on this data sheet.
https://earthworksaudio.com/wp-content/uploads/2017/07/ZDT-Data-Sheet-2016.pdf

https://earthworksaudio.com/wp-content/uploads/2015/02/EarthworksZDT_web.pdf

The architecture shows gain is set by attenuation up to 60 db worth. Then 20 db on top of that is available as actual gain on the output. I think the first part has a fixed dynamic range or close to it, and the latter can amplify gain and noise so you don't get additional dynamic range with the high output levels.

 

[AnalogSteph]

The part I'm not sure is right is maximum input level minus EIN the way you are doing it. The measurement would indicate 30 dbu and 130 dbu lower for the EIN. A dynamic range at a given setting of 130 db. The 160 db would only seem the case considering the lowest level at low gain and the highest level at high gain. I believe you'll have a 130 db wide dynamic window however like my Antelope interface. I've not tested bunches of mic pres so maybe I have it wrong.

I'm fairly sure I have this right. The math isn't that complicated.

There is additional info on this data sheet.
https://earthworksaudio.com/wp-content/uploads/2017/07/ZDT-Data-Sheet-2016.pdf

https://earthworksaudio.com/wp-content/uploads/2015/02/EarthworksZDT_web.pdf

Oops. I must have had a bit of a brain fart earlier, I have no idea why I was assuming that minimum gain is 0 dB (it's clearly +5 dB, with no EIN spec, though assuming close to the same -110 dBu(A) output noise as for +20 dB shouldn't be too far off I imagine).

So the -132 dBV(A) = -130 dBu(A) shorted EIN actually applies to +20 dB, for a dynamic range of 140 dB(A) ref. +30 dBu out.
It may make it the odd dB beyond that at lower gains but probably not too much more, and I imagine distortion performance would eventually be degraded by common-mode effects. If output noise remains close to -110 dBu(A) at minimum gain, EIN would be -115 dBu(A) then... still seems adequate for the Earthworks mic. The ADI-2 Pro would cover the entire range via its various input ref level settings. If @pozz needs more than any of the available 123-124 dB(A) windows, well, he could invest into an XLR splitter cable (ca. +2 dB) and maybe even go composite ADC with a second ADI-2 FS (@ +4 dBu while the ADI-2 Pro is at +24 dBu, wordclock sync obviously required, manual crossfading in audio editor or something). It wouldn't have to be an ADI-2 FS even, most any audio interface with wordclock sync and a good mic input should work, just keep an eye on recovery time after clipping. (*) Or perhaps run both ADI-2 Pro channels at +4 dBu and use an external 20 dB inline attenuator for one?

140 dB(A) still is very high but not actually implausible, supplies also are high after all. That's about 2.5 µV(A) worth of output noise at +30 dBu. Still a good bit ahead of the HV-3C.

So my math was actually correct but it doesn't help if you plug in the wrong numbers. I'm a bit prone to glitches resulting in inexplicable utter nonsense at times, I'm afraid - I basically need Bruno Putzeys level negative feedback to function correctly.

The architecture shows gain is set by attenuation up to 60 db worth. Then 20 db on top of that is available as actual gain on the output. I think the first part has a fixed dynamic range or close to it, and the latter can amplify gain and noise so you don't get additional dynamic range with the high output levels.

That 20 dB is variable attenuation for the variable output, it is not relevant when looking at the main signal path. Said variable output is intended for better matching to inputs that can handle far less than +30 dBu, e.g. +8 dBu consumer line level or whatever.

*) Independent input ref level for both channels might be a good feature request for the next iteration of the ADI-2 Pro. That would make these composite ADC experiments a great deal easier. I'm not sure whether it would be wise to ask for composite ADC out of the box, getting it right seems kind of involved with no shortage of coding man-hours, and it may unduly degrade input latency though I don't think that would be too critical (you could monitor just the top ADC before its signal goes into composite processing).