Don’t know if it gives anything away, but RAW data taken from the input of new ADC is 8 parallell bits every masterclock
So a substential datarate, 2 bits pr input phase each masterclock period
So maybe time multiplex more channels like the mcu sar ADCs
Or downsampling and averaging for mid + lf + unchanged hf. Fir first and joining after.
Things that is done im mcus today for normal signal conditioning
Or downsampling and averaging for mid + lf + unchanged hf. Fir first and joining after.
Things that is done im mcus today for normal signal conditioning
By the way, a cascaded ADC setup could be completely software-implemented. Left channel 1.7V, and the Right is 10V, other things are SW.
So just go fiddle with the ADCiso dipswitches and write the code. (Kind of a joke, but it is maybe possible?)
I posted that txt in Audacity forum/feature request:
The Cascaded ADC is a well-known technique used for 40 years or so in industrial instrumentation. The idea is simple: you have 2 poor ADCs, and you set different gains for each one. A high-gain ADC converts small signals, a low-gain one handles max possible signals. In terms of Signal/Noise, the resulting Cascaded 2 ADCs perform as a better single ADC. A few years ago, such an idea was implemented for audio recording as well, again 2pcs cheap ADCs + DSP to switch converters online. As a result commercial product boasts “phenomenal dynamic range” almost for free. I think Audacity may offer that for a stereo ADC(Lch high-gain, Rch is low-gain etc). An audio signal requires a bit special switching algorithm around the gain split point to hide audible artifacts, similar to a compressor’s attack/release. I believe, we need implement an instant switching action in case the high-gain ADC is overloaded, just to avoid clipping. If the signal drops under the split point for 200-300mS we do switch back from the low-gain ADC to high-gain one(probably with a linear interpolation or fadeout to reduce artifacts if any). The gain split point should be found during the calibration procedure(0dbfs 1kHz for a second or so).
A practical example: my 126db(A) of DR ADC has 2 inputs with independent gain settings from 1.7V to 10V(AUX input from 35V to 43V but it is too much I think) i.e. 15db range with inputs >10V immunity. Hence, with my ADC + Audacity Cascaded ADC feature, I can get 126db+15db = 141db(A) of the resulting Dynamic Range. In other words, we can use an extremely high DR condenser mic 130-140db to record drums or whatever a jazz vocal with the same ADC and preamp settings.
Of course, I understand that the Cascaded ADC feature for audio recording is rather a marketing idea than a real improvement, but it is a sexy and unique feature, especially if 100.00% SW is implemented.
What do you think?
The Cascaded ADC is a well-known technique used for 40 years or so in industrial instrumentation. The idea is simple: you have 2 poor ADCs, and you set different gains for each one. A high-gain ADC converts small signals, a low-gain one handles max possible signals. In terms of Signal/Noise, the resulting Cascaded 2 ADCs perform as a better single ADC. A few years ago, such an idea was implemented for audio recording as well, again 2pcs cheap ADCs + DSP to switch converters online. As a result commercial product boasts “phenomenal dynamic range” almost for free. I think Audacity may offer that for a stereo ADC(Lch high-gain, Rch is low-gain etc). An audio signal requires a bit special switching algorithm around the gain split point to hide audible artifacts, similar to a compressor’s attack/release. I believe, we need implement an instant switching action in case the high-gain ADC is overloaded, just to avoid clipping. If the signal drops under the split point for 200-300mS we do switch back from the low-gain ADC to high-gain one(probably with a linear interpolation or fadeout to reduce artifacts if any). The gain split point should be found during the calibration procedure(0dbfs 1kHz for a second or so).
A practical example: my 126db(A) of DR ADC has 2 inputs with independent gain settings from 1.7V to 10V(AUX input from 35V to 43V but it is too much I think) i.e. 15db range with inputs >10V immunity. Hence, with my ADC + Audacity Cascaded ADC feature, I can get 126db+15db = 141db(A) of the resulting Dynamic Range. In other words, we can use an extremely high DR condenser mic 130-140db to record drums or whatever a jazz vocal with the same ADC and preamp settings.
Of course, I understand that the Cascaded ADC feature for audio recording is rather a marketing idea than a real improvement, but it is a sexy and unique feature, especially if 100.00% SW is implemented.
What do you think?
It's not that simple, there are a few things wrt hardware to consider, based on some investigations I've done plus first experiments with real hardware:
- DC offset and drifts and AC gain difference and drift must be handled. The reference voltage for the ADC sections must be one and the same source.
- any DC-filter / AC-coupling / LF-filter must be in the common path before the split.
- any lower frequency anti-aliasing lowpass, like at 30kHz or so, must also be before the split.
- neither the ADC itself nor any opamps must run into uncontrolled clipping throwing off operation points, that is, one needs a nice analog precision soft-clipper in the high gain path so the recovery from the clipping is instant without any hangover or any caps taking time to re-charge to correct values (notably in any integrators / low-pass-filters when the opamp went open-loop).
- instant switching does not really work, rather one needs to track the voltages on both ADCs and once the high gain path approaches its clipping levels (at 1/5th or so of this point) then a weighted average must be formed, implementing a crossfade.
- for the whole things to reliably work, the sample-rate should be higher, like 2x the lowpass filter frequency.
- so far I've not found that state would be really needed, like a time-based hold-off once the low-gain path was fully activated but the levels have fallen below thresholds. Further experimentation is required....
For the stepped response it would actually be nice to be able to switch from low gain to high gain in one run. Because the measurement can take hours.
Maybe something for REW.
To runs.
Automatic shift between left and right channels between them.
Maybe something for REW.
To runs.
Automatic shift between left and right channels between them.
I dunno what I'm doing wrong, but the most trivial Matlab code makes a smooth result, I can't hear any artifacts. Cosmos ADCiso Lch 1.7V, Rch 10V.
https://drive.google.com/file/d/1swmW-jH56qJhCk4iFnBLahqW2LbBT1Qs/view?usp=sharing
https://drive.google.com/file/d/1swmW-jH56qJhCk4iFnBLahqW2LbBT1Qs/view?usp=sharing
Like the way you marked the transitions!
The wrong thing must be to only use one sample for the ratio calculation. That must be worst case for noise😉
So the max error is max value of analog pop or click noise.
But that should be max 10 dB worse than rms. So maybe max -115dB. No wonder you can't hear it.
Actually I think the ongoing ratio calculation was quite smart!
The wrong thing must be to only use one sample for the ratio calculation. That must be worst case for noise😉
So the max error is max value of analog pop or click noise.
But that should be max 10 dB worse than rms. So maybe max -115dB. No wonder you can't hear it.
Actually I think the ongoing ratio calculation was quite smart!
With ongoing ratio calculation, I would prefer a little averaging, drift in analog components are just measured and compensated. Smart.
Wonder how that ratio changes over time?
A long, long time ago I studied MP2. MP3 predecesor, kindof. In precence of signal the ear masks changes in quantization noise (Or just noise) very well.
So if the noise changed 50dB under signal, maybe there would be artifacts, but at -100dB, no way in my opinion
Just for fun: NICAM is not that far away from this but has more than 2 levels but lack som bits compared to the 32 of today: https://downloads.bbc.co.uk/rd/pubs/reports/1978-26.pdf
Wonder how that ratio changes over time?
A long, long time ago I studied MP2. MP3 predecesor, kindof. In precence of signal the ear masks changes in quantization noise (Or just noise) very well.
So if the noise changed 50dB under signal, maybe there would be artifacts, but at -100dB, no way in my opinion
Just for fun: NICAM is not that far away from this but has more than 2 levels but lack som bits compared to the 32 of today: https://downloads.bbc.co.uk/rd/pubs/reports/1978-26.pdf
Last edited:
All this just for fun, I'm not a believer if cascaded ADC anybody needs for audio, excluding marketing guys. Industrial instrumentation, maybe.
What about this? The result is the same smooth, ES9822 noise is probably too low to affect the process.