Thanks for this, it adds additional bits.
One of more interesting sentences about TPTB:
"Stuart Yaniger suggested that we forsake the trendy, high performance audio operational amplifiers like the OPA637 or 49720 and choose a line amplifier with more humble power supply performance."
In a linestage this may very well be true. I would probably have suggested Borbeleys non-nfb line stage.
Instead they apparently went with this:
Now exactly why anyone would consider this circuit to have poor PSRR is beyond me, it will do quite horrorshow. I can nly estilate, but with the K170/J74 being in the region of 50K+ as CCS and more in this circuit it has at least 20dB inherent PSRR, probably more.
The Amplifier has around 60dB open loop gain, if we take this with a linestage gain at 12dB, we should have around 68dB PSRR up to around 5KHz or so, not bad at all I would say. It would have been good if this test would have actually supplied measured data for the linestage.
We also know that the PSU usd 4,700uF/0.68R/4,700uF per rail with maybe 50mA total current consumption.
PSUD predicts 26mV Peak-Peak ripple with this arrangement, I'd expect 10mV RMS, so if we were not using any regulator at all, I'd expect 5uV 100Hz and much less higher up. To put this into perspective, we are talking here about around -106dBV 100Hz noise.
I would suggest that in fact parameters like self noise and PSRR of any of the regulators in the selection put things well beyond "good and bad", any regulator that can manage 40dB ripple rejection would drop PSU noise below 24Bit levels and way below the circuits self noise.
So in effect, any differences between the regulators would be down to supply current modulation and the resulting error voltages feeding back into the circuit. I would further add that with the presence of local 220uF decoupling capacitors most of the differences heard could be simply down to interactions between these cap's and the regulator and dynamic current demands.
It is interesting to note that the output impedances of the top rated regulators vary quite widely, however three of the top five have very low output impedance, with only the Burson regulator showing a quite high impedance. The Linear Tech regs coming next also have high output impedance.
To me this suggests that the real factor affecting sound quality was not sufficiently captured in the test.
I would like to offer a hypothesis though. The source used was an SACD Player. These are known for prodigious levels of supersonic and low RFI noise on their output. The Borbeley Line stage has very wide bandwidth, but was decoupled only with 220uF electrolytics, which are likely next to useless above around 100KHz (see Bateman), so at very high frequencies within the operating bandwidth of linestage the circuit relies primarily on the regulator for rail impedance and this will be all over.
This will be made worse by the fact that the following Amplifier is also quite wideband, the interconnects offer minimal if any RFI filtering themselves and that the following Amplifier is comparably non-linear, not grossly so, but certainly more than one may wish if one has a lot of RFI riding down the line.
Clearly, two regulator designs will show themselves superior in this case, for one those that start out with extremely low impedance (Jung/Didden) as they will still have comparably low rail impedances at high frequencies, the others will be very wideband designs, which I believe applies to the Burson and Linear Tech designs.
Only a hypothesis and one that requires taking into account wide ranging factors, however, the perhaps crucial lesson is that we must account for a wide range of factors in such tests and even more so in designing gear.