Recipe: how to cook a best in class 2-ch pre-amp with a/d convertor?

[laurentboutonnat]

After much discussion in another part of this forum I thought this to be of such an importance that I wanted to share this with you. I have saved myself around $7000 with this information and the sound is unpreceded. The boutique companies are selling you old convertor technology, although of course they want you to believe their technology is 'special'.

The pcm4222evm will deliver the best a/d convertor technology avaible at the moment, because it uses a new chip (2006), extremely quite and fast op-amps (OPA1632) and directly coupled. Did I mention the whole circuit is differential....,yes?

First of all: this recipe is not meant for people who have never had any physics classes or ever touched a soldering iron before! It can be even dangerous when working with 110V/230V without proper knowledge of grounding and safety measures.

Ok, here's the recipe:
Step 1.You take a nice box like the ready made SC-1 rackcase from Five Fish or 'gut' an old Behringer you find in your garage. ($$?)

You purchase two ready built +18/-18/+48V and +5/-5/+5V psu's from linear technologies or build it yourself using the PSU 1884 kits and a torriodal transformer.

Step 2. You buy the PCM4222EVM from TI ($150)

Step 3. You buy one or two SC-1 pre-amp kits from Five Fish which will give you better results than the DAV BG-1, which uses the old INA217, while the SC-1 uses the much better THAT1512. ($99)

STEP 4. Connect it altogether and use the AES3 or S/PDIF interface to get the digital signal in your computer, don't worry about jitter because that's not a great problem anymore. Although some manufactures will make you believe otherwise (yeah, it's all in the marketing)... If you use a post 2000 product with a decent receiver chip, which reclocks all information anyways, you have no problems at all. Of course this is only the case when a decent 75ohm cable is used for S/PDIF and a 110ohm for AES and the cables are not longer than 10 meter. The 4222evm itself has an extremely low jitter cmos crystal clock generators. For the real die hards it's even possible to output the signal as a 1-bit DSD signal and record it using products supporting one bit technology. But the PCM output will give you 123 dB at ease..

Don't forget to add salt and pepper and you're ready to record with the cleanest and best technology avaible today!

If you don't want a clean, warm, full sound you can later always color and degrade the sound a bit when necessary using simulation plugins like the Waves API series or just run it through a compressor with lots of iron in it.

You have everything you need in terms of amplication. Microphones is a completely different story, which I will tackle another time.....

[telecasterrok]

very interesting!.. how could one record, say 8 channels at once? Is there a way to hook them all into a sound card at the same time? Thanks for the tip!

[lespaul666]

If you made 4 of the units with double-channels and have soundcard/interface with 8 AES inputs, or a format converter to ADAT lightpipe, it would all work fine for 8 channels.

[Francis Vaughan]

You need to provide an external clock to the boards if you do this. This will get a bit more messy. It isn't a word clock either, but must be one of 22.579 or 24.576MHz. Don't underestimate the pitfalls in getting that right.

[telecasterrok]

Okay... So I'll need an external clock source to do more than 2 channels. Is there an affordable clock on the market that will work with these chips? Thanks for your help!

[laurentboutonnat]

It should not be difficult to let the 4222 output a good worldclock signal. Since the board has holes on all important places in the circuit, soldering two wires should be enough. Let me work on it...I will report back to you soon.

[Francis Vaughan]

Moving a wordclock around is fraught with all sorts of problems. You are working in the RF domain, and the issues with impedance matching, transmission lines, reflections and terminations are really awful. Lots of care is needed. It is easy to get something that works, but even easier to get something that has very difficult to diagnose and sort out sonic problems. To manage 16 bit accuracy you need to keep the phase noise under .3ns, and for 20 bit, 10's of picoseconds.

The best and easiest way to do this is to keep all the ADC chips on one board, and use stripline construction to distribute the clock. A 6 layer board is pretty much mandatory. If you use coaxial connections and a distribution amp typically a time domain reflectometer is needed to ensure a good distribution.

It may be that the eval board's design, where it doesn't take a word clock, but takes and external crystal frequency clock may be slightly easier to cope with, but in reality the same care is needed.

[boxymoron]

I just recieved my pmc4222 eval board on the mail. I also ordered a power supply kit for the +/- 15v and +5v lines, and a digital coax cable. Next will be assembling the power supply and building a nice rack enclosure from par-metal, which is actually about 10 miles away.

Btw, I was looking at an application note from maxim semiconductor which provides details for a 6nV/sqrt(Hz) noise density performance regulator using a high precision opamp and voltage reference. The only problem is that the voltage reference chip comes in smd only (I think), which makes it almost impossible to diy. I'm also thinking about adding some kind of metering at the inputs,to monitor levels, since the eval board doesn't have any kind of clipping protection. Eventually I would like to implement digital decimation and low-pass filtering with DSP, along with a digital limiting/soft saturation algorithm using the 6-bit outputs from the board and a DSP eveluation board.

[laurentboutonnat]

Some very ambitious plans boxymoron, but I like it! Maybe it would also be very interesting to record the 1-bit output directly and do the decimation in software. Which pre-amp are you planning to use? What do you think are the benefits of using a FPGA for the decimation filtering instead of the on-chip filtering?

Please keep us informed about your progress....

[boxymoron]

There's nothing wrong with the built in decimation filter. It seems to me that the best place to implement some kind of limiting process would be after the modulator output, when the signal is still "oversampled". It is in this realm that non-linear processing (limiting/compression) can be accomplished and artifacts from such processing be pushed outside the audible band and filtered out afterwards (it's mostly high frequency content/harmonics, so we use low pass filtering). Of course, we are not really "limiting" the input(analog) signal, since once the maximum amplitude of the a/d input is exceeded we have "lost" information, that is, you can't go above "1" for ONE SAMPLE, for a 1-bit converter, but because the signal is OVERSAMPLED we can "bring it (the signal at some arbitrary frequencies) down", while pushing the artifacts (other arbitrary frequencies) of such processing outside the audible band.

Mental experiment: Suppose we are sampling ONE signal with two 24bit a/d's. One has a maximum input level of 4V p-p (in other words, it's referenced to 4V p-p) , and the other to 8V p-p (or 16/32/64/128...). When the input signal (at one sampling instant) exceeds 4V p-p, the second converter can still output a "correct" sample value for that sample. However, we are interested in outputting a digital stream that represents/corresponds to a reference voltage of 4V p-p(analog). So, we use the "meta data" afforded by the second converter to process the data from the first converter and make sure it fits into a 24bit/referenced-to-4Vp-p signal.

Oversampling limiting allows for a mechanism similar to this (in the frequency domain), because the signal is OVER sampled, we have the necessary "meta data" to work with.

This, I think, is best accomplished with very high sampling rates (128-256) of the original signal such as the output of an a/d converter, as opposed to upsampling an already decimated/filtered signal. It would require lots of processing power to do this kind of processing on a MHz range signal, which is why an FPGA or very fast DSP processor would be required. We are talking probably in the order of billions of computation per second!!

Maybe someone with a deeper understanding/knowledge of the problem can chime in, as I seem to be stuck thinking in terms of time-domain signals at this "instant". ;^)

[laurentboutonnat]

Hi, I am missing something here. What's the benefit of limiting after A/D conversion when it would be much more easy to just do it before the conversion? Just put a peak limiter between the pre-amp and the convertor. Just calibrate it and make sure the signal never exceeds the maximum imput of the A/D. Of course you want as less limiting as possible, so you make sure (almost) no limiting is going on when setting gain of the pre-amp. Besides that you have so much headroom that driving the ADC a little sweeter to prevent clipping, would give a dynamic range of 115db> easily. If you record to put it on a cd why bothering about DNR when it becomes 90db eventually anyway. Besides that most music recorded today not even uses 75% of this full 90db of DNR, because many engineers love to use their Waves L2 and alikes. When your song is played on the radio through an Optimod, with extremely agressive multiband limiting/compression and even clipping, the DNR is reduced to about 40 db! So why do we bother about convertors' dynamic range anyway....?

[boxymoron]

Because many, including high-end mastering studios, use their a/d converters to compress and limit tracks. I'm guessing there's a difference btw digitally limiting an "original oversampled signal at the a/d outputs" than limiting an already decimated signal that has been upsampled (interpolated).

[craigmorris74]

Because many, including high-end mastering studios, use their a/d converters to compress and limit tracks

What converters do this well? I've never heard of this practice.

[laurentboutonnat]

I think it's easier to record the 1-bit stream, compress/limit this stream using software capble of doing it (Pyramix) and than decimate if neccesary...all you have to do is get the I2S stream in your computer using an usb front-end, works perfect.

[danasti]

Originally Posted by craigmorris74

What converters do this well? I've never heard of this practice.

Lavry's Gold - you've maybe never heard of it but I'm sure you've heard it. It's done quite a bit at Sterling. I think it can sound great... Not sure if the limiting is done pre or post conversion on that converter.. Any info?

[KSwan]

you know... this sounds like an awesome project. If you've really in fact done it... please please post a sound sample or two.

[boxymoron]

Still waiting on the power supply. Will have to do a lot more research before I can implement a digital limiter. Haven't found any documentation on 6-bit modulator outputs.

[KSwan]

What about you, laurentboutonnat? Don't you have any sound samples to share?

[laurentboutonnat]

Coming! Stay tuned...

[boxymoron]

First test. This is really testing BOTH the D/A in my Yamaha I88X and the A/D in the PCM4222 evaluation board. The test setup was as follows:

• Nothing was calibrated.
• 16-bit 0dBFS-peak-amplitude sine wave .wav file (courtesy of RME Audio)
• D/A output is direct from a I88X output opamp (ie it has no volume pot or attenuator)It's rated at +18dBu max output level.
• A/D input is direct into the A/D opamp buffering stage(ie no preamp, amplification factor is -.482(-6.34dB), A/D is rated at +14.68dBu for 0dBFS)
• Unbalanced and crappy (less than optimal - lose 6dB) cable interconnect
• Recorded sample was normalized to 0dBFS (since the D/A and A/D have different reference levels).
• Recorded Sample was "time aligned" to original sample and then phase reversed (does not include all latencies involved)

So, you can see the test is pretty flawed to start with. I am really measuring both the A/D and D/A. When summed together the waveforms don't cancel out 100%. However, looking at the waveform on my audio editor (zoomed in at the sample level) they "look" almost identical. BTW, I tried clipping the A/D with the Yamaha and wasn't able under normal circumstances. The only way I could (and only barely) was driving it with the master outputs in the I88X, which do have an active output stage (volume control). I guess this is because of the 6dB drop from running unbalanced cables. The original file is named 0_16.wav

I'll make some other tests, including an actual music recording...

[boxymoron]

Of course, the proof is in the pudding. I will post some ausio/music samples as soon as I can get a proper setup.... Stay tuned

[boxymoron]

Another non-musical test. This time its the crest factor sample. Crest factor is the difference between the peak and average value of a signal, in this case peak is 0dBFS and rms is -16.1dBFS. This time I adjusted the recording offset (146 samples) so its prefectly in sync and normalized to 0dBFS. Looks pretty close to me!

[Rochey]

Good to see things are coming along in this thread.

Keep up the good work

[KSwan]

Yes, those clips do sound promising. Thanks for posting them. It would be nice to hear something more musical, too, though.

[boxymoron]

Finally managed to record something "musical", if you can call it that, . Comparison files are at: Welcome To Vacuum Tube Central® under A/D Converter Tests section. The quality is very poor due to the cheap microphones used, poor mic placement, poor performance, poor acoustics, etc.

[boxymoron]

For those interested, here are a couple of links to archived prosoundweb discussions on the topic of impulse/transient response in band-limited systems (mostly layman's terms discussions):
PSW Recording Forums: Dan Lavry => micro-timing in low sample rates
PSW Recording Forums: Dan Lavry => the "high frequency transients" fallacy

And for those more technically inclined,
Continuous-Time Sigma-Delta A/D ... - Google Book Search

[laurentboutonnat]

First musical test recorded as fair as possible.

I use the output of a raw multiband recording using a switcher into the different ADCs which are recording at the same time.

You can hear which ADC is running when.

Units used:
1. PCM4222EVM
2. Lavry Gold
3.Fireface (just for fun)

Here's the test.

My opinion: I think the PCM and Lavry are really close. It's just a matter of taste what to chose. I think the PCM has a bit more DNR and is a little more precise and colder (due to the faster op-amp). The Lavry sounds a bit warmer, but also less precise.

[boxymoron]

Originally Posted by laurentboutonnat

First musical test recorded as fair as possible.

I use the output of a raw multiband recording using a switcher into the different ADCs which are recording at the same time.

You can hear which ADC is running when.

Units used:
1. PCM4222EVM
2. Lavry Gold
3.Fireface (just for fun)

Here's the test.

My opinion: I think the PCM and Lavry are really close. It's just a matter of taste what to chose. I think the PCM has a bit more DNR and is a little more precise and colder (due to the faster op-amp). The Lavry sounds a bit warmer, but also less precise.

Thanks for the clip Laurent,
I can clearly hear the differences in this test (compared to mine). To me it sounds like the Lavry and PCM4222 have a little more "weight" in the bass region than the Fireface and sound a little more dynamic too. On the other hand the Fireface and the Lavry sound a little bit warmer in the high end than the PCM4222, which sounds brighter and more detailed (a bit sterile some would say). The mids are more difficult to compare in my crappy speakers. I can also detect a little "sibilance" in the esses with the PCM4222.

Did you calibrate all three to the same reference levels, or do any kind of normalizing? I ask because a difference of even 1dB can make a difference in percieved sound quality. Personally, I think a test like this is somewhat compromised (in terms of the source signal) because you're also listening to the source's D/A conversion stage. Also, the source is mp3 encoded and you are listening to different parts of the songs (not very objective in my view). Still, it's a nice test. I'm thinking a nice tape machine feeding the converters (calibrated) simultaneously at the top range (peaks at -1dBfs) would be a more accurate and definitive test.


BTW, I really like the Lavry's "warm" coloration, but is it really worth 50 times more than the PCM4222?

[laurentboutonnat]

You're right. I will fix some issues with the test and will report back soon. I want this to be as scientific as possible, it's difficult but possible. Differences between Lavry and PCM are really subtle. I think the Lavry uses a tranny, the pcm4222 not. That would explain the difference in warmth. But you could just put a tranny before it or use a plug-in like the URS series (I will try this). Of course the Fireface is in a different league...performance wise.

[DeeDrive]

Originally Posted by laurentboutonnat

First musical test recorded as fair as possible.

I use the output of a raw multiband recording using a switcher into the different ADCs which are recording at the same time.

You can hear which ADC is running when.

Units used:
1. PCM4222EVM
2. Lavry Gold
3.Fireface (just for fun)

Here's the test.

My opinion: I think the PCM and Lavry are really close. It's just a matter of taste what to chose. I think the PCM has a bit more DNR and is a little more precise and colder (due to the faster op-amp). The Lavry sounds a bit warmer, but also less precise.

Hey, I think the link is dead, can you repost it? Thanks for doing this, I'm thinking about sticking on of these boards in a 2 channel API 312 clone, mmmm....