Power rail conducted SMPS noise may be the easier part to think about and plan for. Its the sneak circuits you might not see that can get you, radiated noise, noisy ground currents, etc. Those sorts of things can affect the sound even though measurements look acceptable.
PCB layout and ground planes matter, but if measurements wouldn't still reflect reality then I would pass it, since it falls into the category of black magic, that cannot have a scientific explanation. Besides, sound quality is a subjective thing anyway.
So, once again, that means that one could do his best version of everything (power supplies, filtering, PCB or whatever) and leave the rest to nature.
So, once again, that means that one could do his best version of everything (power supplies, filtering, PCB or whatever) and leave the rest to nature.
Mumetal box, then copper clad pcb box, the. Stainless steel box. Batteries inside for power.
That’s how Jim made his low noise amp and i suspect the same level of complexity to reduce environmental noise on the traces etc for -130dB or below..
For the front end op amp you may want to look at matched resistor network opamps (like the LT1991 or an equivilent).
I’m on the same journey but i have a prebuilt AKM 5572EN board. Not as low dB as yours but that’s bad enough.
The clock i was considering a thermal OCVX (if i have those letters in the right order), and thinking a super cap just for the clock power supply. So the cap is charged and the supply then covers the period of the sampling.
I was thinking a small linear 3042s for the PS with filtering before it. Possibly with a fast RF opamp shunt regulator.
However those are future ideas - i’m looking at usb-i2s at the moment using a baremetal rpi zero to act as large fifo buffer. Without the interrupt conflict of the rest if raspberry OS, it’s DMA should keep up easily.
Last edited:
I2s and rpi. At high speed may render the RPI unusable if it has to also run the user’s desktop and applications.
You may find the number of interrupt conflicts resulting in queues may cause issues. The RPi only has a 64 slot 16bit FIFO for the pcm/i2s IIRC. That’s 64 is then spread for both channels and sample size.
My thinking here is a bare rpi can muster enough bus bandwidth and interrupts per second to handle a FIFO 20 sample FIFO threshold to trigger and dma through interupt into a ring buffer which can then package into larger 1024byte packets for the USB iso transfers. Thus minimising chance of drop out via underrun/overrun and latency being what USB will deliver.
An option may be store to memory ot sd card and then access via filing system bulk transfer if it’s that critical.
You may find the number of interrupt conflicts resulting in queues may cause issues. The RPi only has a 64 slot 16bit FIFO for the pcm/i2s IIRC. That’s 64 is then spread for both channels and sample size.
My thinking here is a bare rpi can muster enough bus bandwidth and interrupts per second to handle a FIFO 20 sample FIFO threshold to trigger and dma through interupt into a ring buffer which can then package into larger 1024byte packets for the USB iso transfers. Thus minimising chance of drop out via underrun/overrun and latency being what USB will deliver.
An option may be store to memory ot sd card and then access via filing system bulk transfer if it’s that critical.
Really? In order to help out the OP here may we ask which measurements would you suggest the OP perform in order to cover all possibilities presently known to science?
Last edited:
That, I believe. I cleaned up emissions from an SMPS once using a portable wide-band radio with external probe antenna to measure. The SMPS was in a sealed metal box lined with copper foil on the inside and fully sealed in copper foil on the outside. Power going in and out was cleaned up with ferrite chokes inside the box. It worked, but wasn't really worth all the effort.
BTW, now I sometimes use a spectrum analyzer and E-field probe. Its quicker than the radio. Maybe not as sensitive though.
Also, you don't really need an oven-controlled crystal oscillator for audio. The oven is for longer term frequency stability which doesn't matter for this. Close-in phase noise is more the issue (at below 10Hz offset from the carrier, 12kHz offset phase noise numbers are for communications stuff not audio). A buffered Crystek 957 with 805 SMD Rubycon .22uf film cap for bypass can be pretty darn good (although better is possible). Small film cap works better than a super cap. Don't use a ferrite bead if imaging is important (hysteresis distortion creates its own noise).
Last edited:
Why should you need to cover all possibilities presently known to science? This is just a RPI ADC.
No. Particular measurements can find particular problems. If we only fix the problems we happen to measure, in so doing we may be worsening other problems we didn't measure. That's the larger problem, as I have pointed out many times. Why can't you understand WYSIATI? https://facilethings.com/blog/en/what-you-see-is-all-there-is
WYSIATI is a scientific fact. The term 'reasonable doubt' is taken from the field of law. Legal facts and scientific facts are two entirely different things that serve different purposes. Whatever a jury of 12 people off the street happens to believe is enough for law. It isn't enough for science.
WYSIATI is not a scientific fact. And anyhow what does science have to do with a RPI ADC? If a set of measurement show no problem then beyond a reasonable doubt there is no problem.
Behavioral science is not a "hard" science that produces facts. There are people who behave as WYSIATI predicts but also people who don't.
Behavioral science is not a "hard" science that produces facts. There are people who behave as WYSIATI predicts but also people who don't.
Next you'll telling us the Earth is flat
WYSIATI is based on a lifetime of world class scientific research. https://www.apa.org/monitor/2012/02/conclusions