• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Best spec ADC Chip currently.... ??

IVX

Major Contributor
Joined
Mar 3, 2019
Messages
1,439
Likes
2,830
Location
South of China, SHZ area, - Слава Україні
I met an interesting dilemma, regardless of great crosstalk figures I still can see how one channel signal affects THD of another one. So, I can compensate THD for 2ch at the same time or independently for each channel(when the input signal at only a single channel). The second compensation for sure has better precision. I ask myself, should I improve that, more and more polish the PCB layout, or just prepare the warning about maximal precision available in the mono-mode?
 
Last edited:

Lightrock

Member
Joined
Mar 17, 2021
Messages
15
Likes
18
Location
Switzerland
To polish the PCB to avoid any crosstalk is certainly the better option. And 4 layers are the absolute minimum to achieve this - 6 would be better to include two solid copper planes. Bear in mind that the crosstalk is frequency-dependent - if it´s capacitive, it´s rising with 6 dB/oct, whereas the distortion compensation scheme in the ES98xx is frequency-independent. So what might look good at 1 kHz might not work out at 2 kHz or 10 kHz.
 
Last edited:

IVX

Major Contributor
Joined
Mar 3, 2019
Messages
1,439
Likes
2,830
Location
South of China, SHZ area, - Слава Україні
Lightrock, did you see my video E1DA Cosmos V03 crosstalk? Did you notice some paradoxical harmonics behavior yet? The crosstalk isn't symmetrical the bottom ch affects the upper one more than vice versa(-135/-140db), however, the bottom channel's harmonics chanced a lot more if the top channel On/Off. I need to realize that first before refining the PCB layout.
BTW, 2 layers PCB proto + 1 jumper had the same crosstalk as 4 layers without a jumper.
 

HpW

Member
Technical Expert
Joined
May 13, 2018
Messages
42
Likes
9
Location
Guacimara
Lightrock, did you see my video E1DA Cosmos V03 crosstalk? Did you notice some paradoxical harmonics behavior yet? The crosstalk isn't symmetrical the bottom ch affects the upper one more than vice versa(-135/-140db), however, the bottom channel's harmonics chanced a lot more if the top channel On/Off. I need to realize that first before refining the PCB layout.
BTW, 2 layers PCB proto + 1 jumper had the same crosstalk as 4 layers without a jumper.

Dual mono would be the preferred solution, otherwise you deal with chip cross talk even on OPAMP's :D

Also to get ride of the hum o_O is one task more :p as with none balanced :eek:

Hp
 

capslock

Senior Member
Joined
Jun 19, 2020
Messages
365
Likes
184
Ivan, so what you are saying is that you get lower HD for at least one channel as opposed to adding both channels, whereas you get lower noise from adding both channels?

For me, I always prefer low harmonics, noise at that level does not bother me as I don't change DR.

The ES9822 has diff input, so you have a switched single ended preamp and then you use an inverter to generate a diff signal? You could end up with the same op amp count if you were to stay diff, but as you have hinted, this would mean adding switches and resistors. I don't trust MOSFET switches because of voltage dependent capacitance. Two pole relays?

Adding a diff to SE converter will not use up too much space, but if it is not switched, its noise performance won't be too good.

How about integrating a USB isolator, and then you no longer need diff in?

I am actually more worried about midlevel distortion that Lightrock mentioned.
 

capslock

Senior Member
Joined
Jun 19, 2020
Messages
365
Likes
184
Frankly, I find those videos hard to understand. There are no subtitles explaining what changes between measurements, no summaries and it's all very fast.

Is your comment about adjusting HD for each channel vs. for both? Or about midlevel distortion? Midlevel is not -3DBfs.

Just coming to think of it, if you can adjust each channel separately for minimal HD, then the sum would still be better than if you had adjusted an average value for both. I can see no advantage whatsoever to adjusting for both at the same time.
 

Tks

Major Contributor
Joined
Apr 1, 2019
Messages
3,221
Likes
5,504
If somebody still interested, I've got THD+N -119db/-121db stereo/mono mode with ES9822.

You happy with that, or you do you want to squeeze the chip for every ounce of life it has ;P
 

chris719

Senior Member
Joined
Mar 22, 2019
Messages
373
Likes
425
If somebody still interested, I've got THD+N -119db/-121db stereo/mono mode with ES9822.

Very nice. How is the bandwidth at higher sample rates? Are there signs of early noise shaping like AK557x?
 

Lightrock

Member
Joined
Mar 17, 2021
Messages
15
Likes
18
Location
Switzerland
If somebody still interested, I've got THD+N -119db/-121db stereo/mono mode with ES9822.
Of course, everyone is still interested! Very nice performance, IVX! Are these numbers achieved with the H2 / H3 overtone compensation activated or bypassed? I get slightly inferior numbers near fullscale (without THD compensation), but I am working with the 4-ch version ES9842.

One thing that really annoys me a lot is that in contrast to the AK5578 (and also the ES98x0), the 3.36 Vrms needed at the differential analog input pins to drive the ADCs to fullscale require that the ADC input buffers are fed with slightly more than the usual 0, +5V. I have provisonally soldered in an old LM337 connected to -15V to feed the V- pin of the ADC buffer opamps with -1.25 V instead of 0 V. The 1.25 V below GND are probably a good deal more than needed, as for the V+, raising the +5V supply to only +5.3 V is already sufficient to get the signal through without rising distortion up to 0 dBFS .

This solution is of course totally energy-ineffcient. It´s just a small example how stupid decisions from a chipmaker can make the practical implementation unnecessarily difficult.
 

chris719

Senior Member
Joined
Mar 22, 2019
Messages
373
Likes
425
Of course, everyone is still interested! Very nice performance, IVX! Are these numbers achieved with the H2 / H3 overtone compensation activated or bypassed? I get slightly inferior numbers near fullscale (without THD compensation), but I am working with the 4-ch version ES9842.

One thing that really annoys me a lot is that in contrast to the AK5578 (and also the ES98x0), the 3.36 Vrms needed at the differential analog input pins to drive the ADCs to fullscale require that the ADC input buffers are fed with slightly more than the usual 0, +5V. I have provisonally soldered in an old LM337 connected to -15V to feed the V- pin of the ADC buffer opamps with -1.25 V instead of 0 V. The 1.25 V below GND are probably a good deal more than needed, as for the V+, raising the +5V supply to only +5.3 V is already sufficient to get the signal through without rising distortion up to 0 dBFS .

This solution is of course totally energy-ineffcient. It´s just a small example how stupid decisions from a chipmaker can make the practical implementation unnecessarily difficult.

It's not uncommon for some high end non-audio ADCs. I've had to use +7.5 / -2.5 rails for some ADC drivers. It is annoying though, yes. There is a TI / old National part that will generate you -0.23 V pretty cheaply and I've used that where I need absolute zero in single supply systems.
 

Lightrock

Member
Joined
Mar 17, 2021
Messages
15
Likes
18
Location
Switzerland
Yes, the LM7705. I had considered it, but the negative voltage it generates is slightly too low (in absolute terms). The other problem is that its charge pump frequency is not synchonizable (as with most small CP ICs). So the switching frequency (in this case, around 92 kHz) would be visable in the spectrum when using higher sampling frequencies.
I have used discrete MOSFET CPs synchronous to the LRCK (or multiple integers) in a couple of other projects, however, on this board, there is not enough space to generate the gate signals from LRCK and BCK (involving some glue logic to provide dead time). The ES98xx has three PWM generators which could perfectly do this job. But to my chagrin, they do not have a common reset, so it is impossible (I have tried everything) to put them into a defined phase relationship to each other.
Just a small example how stupid decisions from a chipmaker can make the practical implementation unnecessarily difficult. :D
 

IVX

Major Contributor
Joined
Mar 3, 2019
Messages
1,439
Likes
2,830
Location
South of China, SHZ area, - Слава Україні
JohnPM, can REW FFT work with 2ch simultaneously? I believe, I saw some screenshots with two FFTs but can't get any clue with google search yet.

The new optimization step brought <-140db@1kHz crosstalk, now I can measure both channels at the same time with no degraded precision.
2021-04-23_21-54-27.jpg
 
Last edited:

KSTR

Major Contributor
Joined
Sep 6, 2018
Messages
2,883
Likes
6,452
Location
Berlin, Germany
@IVX, the standard version of REW is single channel, don't know if the Pro version is multichannel in the RTA.

Workaround: REW can run multiples instances and when you have an ASIO driver that is multi-client capable then you can use both instances at once (at the same sample rate). I haven't tried this with receive data (I have never felt a need for reatime 2ch FFT, what is your use case?) but with send data it works perfectly, even when sending to the same output, the driver simply adds all data together before sending it to the interface.
 
Top Bottom