So, finally, I stopped on something like that:
As you can see, the only "0db" mode requires some postprocess EQ in REW FFT. However, as I mentioned above, that mode needed for the DR test i.e. BW20-20k and A-weighted, hence +.69db at 20k isn't really an issue. 18db peak at 70k perhaps may create trouble with chip DACs like CS43198/131, and better to add one more switch to alternate the termination 10k/1k, or keep 30-50ohm as a min resistor instead of 10ohm(I'm greedy regarding the noise so I'd like 10ohm). The R/r resistor divider steps could be different but I like 10 + 1000;600;400, from a total 5pcs 200ohm MELFs(2.5W together).
As you can see, the only "0db" mode requires some postprocess EQ in REW FFT. However, as I mentioned above, that mode needed for the DR test i.e. BW20-20k and A-weighted, hence +.69db at 20k isn't really an issue. 18db peak at 70k perhaps may create trouble with chip DACs like CS43198/131, and better to add one more switch to alternate the termination 10k/1k, or keep 30-50ohm as a min resistor instead of 10ohm(I'm greedy regarding the noise so I'd like 10ohm). The R/r resistor divider steps could be different but I like 10 + 1000;600;400, from a total 5pcs 200ohm MELFs(2.5W together).
With higher source impedances, you get drooping at high frequencies with R2 & C2 in place. This shows what happens as you change the source impedance from 0 Ohms to 100 Ohms for the zero attenuation (10 Ohms at your R7) low noise setting:
Most of these filter topologies are far more source sensitive than load sensitive. You could fix that by changing the impedance of the voltage dividers at the inputs so that the source impedance is relatively constant. Make R4, R10, and R14 something closer to 10 Ohms and adjust the series resistors to get the division ratios you want. You might have to live with a minimum attenuation to get the flatness correct over a range of source impedances.
As usual, you end up with trade-offs between resistor noise and load impedance. Complex filter and switching or peaking and drooping.
So, would a Scaler handle the out of band signals OK without causing distortion? If yes, then you could put this anti-alias filter between a Scaler and the ADCiso. Make it balanced. One for each channel. That provides a fixed source impedance. No switching needed.
Then, require attenuation before the Scaler for amplifier testing. You already need a high power load for testing amplifiers anyway. Even headphone amps. In fact, with the adjustable gain of the Scaler, you'd only need a fixed attenuator (20 db?) before the Scaler at the highest loss you'd want. It's only two resistors plus the load resistor. Then use Scaler gain to set the gain. That eliminates any switching on the filter assembly and eliminates the need for high power resistors there, too. Plus, isn't the input of a Scaler already AC coupled?
Yes, Scaler and APU AC coupled and have 10u/50V NP Elna, however, I think Cosmos ADC is a lot more popular than Scaler and APU together. Something like Cosmos_ADC/(Scaler+APU) = 5. Cosmos ADC has 220u/10V(ADCiso 16V) polar cap between the 9822 inputs 2.4V biased. Hence, ADC is ok for the DC input range -7.6V to +2.4V.
26Vrms 1k input LPF at -16db setting in fact this is the limit of AP GEN at max possible voltage. After LPF even a bit better. I also tried 2 AP Gens in serial, so 26+26 = 52V, and H3 was the same -123db at the output of LPF about 9Vrms.
Nice work.
If any DIY readers want to make a balanced anti-alias filter that might go between a Scaler and an ADC(iso), here is a candidate. This could also go between a Scaler and an APU before the ADC(iso). Note: Rfix resistors are there to make LTspice happy - they aren't in the actual circuit.
This is what you get for frequency response for all the possible input impedances of the ADC(iso):
If any DIY readers want to make a balanced anti-alias filter that might go between a Scaler and an ADC(iso), here is a candidate. This could also go between a Scaler and an APU before the ADC(iso). Note: Rfix resistors are there to make LTspice happy - they aren't in the actual circuit.
This is what you get for frequency response for all the possible input impedances of the ADC(iso):
Last edited:
When dimensioning this filter - what was your requirement for attenuation at Fs/2... and why that specific figure?
//
//
Thanks - and why that specific figure?
Its about 27 dB down at fs/ 2 for 384 i.e. 192kHz. Is this enough to not end up with significant alias rubbish?
//
Its about 27 dB down at fs/ 2 for 384 i.e. 192kHz. Is this enough to not end up with significant alias rubbish?
//
I just used the same parameters as IVX did, but used the wrong name. My sincerest apologies.
If I could figure out how to delete the post, I would.
Note: Every time I try to be even a little bit helpful around here, in some way, it turns out badly. EVERY time. So, best to learn from that. I'll close the door behind me on my way out..
Please, don't take the DIY forum too seriously. I offered an open discussion of the LPF, and I appreciate your input.
No no... this was no negative critique - but I suppose the level is at times quite high so it is hard around here to get away by just winging it 😉
I just wanted to know... Keep posting by all means!
//
SMSL DL200 -> Cosmos APU Notch 0dB -> Motu UL MK5 ADC
Without the APU and cal file
SMSL DL200 -> Cosmos APU Notch 6dB -> Motu UL MK5 ADC
Without APU
Without the APU and cal file
SMSL DL200 -> Cosmos APU Notch 6dB -> Motu UL MK5 ADC
Without APU
For someone who as an APU, presumably its output is a time-domain notched-out distortion residual? If so, a PC oscilloscope app might be able to visualize it?
back to LPF topic, I think it is not too bad for such simple filter(loaded with APU's Notch inputs) 😉
Loaded by ADC_10V LPF has worse distortions due to higher current through the TDK SMD coil, which should have a high current rating = size(SLF12575T-221M1R3-PF). That ADC sample has a residual H3 at -138db.
Loaded by ADC_10V LPF has worse distortions due to higher current through the TDK SMD coil, which should have a high current rating = size(SLF12575T-221M1R3-PF). That ADC sample has a residual H3 at -138db.
Last edited:
a scope, even 12bits has an awful linearity, and 30db of APU's Notch, for sure isn't enough. Any laptop's internal audio interface is a lot better for that purpose.