Audio analyzers like the Audio Precision APx555 measure the analog output signal of DACs within the 'in-band' range of 20Hz to ~20kHz to ensure they meet design criteria. Signals above this range (~20Khz to several GHz) are 'out-of-band' and not audible and thus not considered of great importance. DACs are rated on their in-band measured performance. However in pursuit of the ultimate experience, audiophiles know that measurements don't tell the whole story; that many tweaks, despite not revealing any measured in-band differences, do improve the sound of a DAC to make it more magical: more transparent and more emotionally engaging.
There are two things going on:
- The human ear/brain is more sensitive than the best in-band audio measurement equipment (which has about -190dB noise floor). Rob Watts (of Chord Electronics) says as much and is pursuing design performance levels over 300dB via simulation (not measurement). Our extreme sensitivity level is a deep mystery but very real. Stop reading right now if you don't believe this (or don't trust your own ears).
- The mechanics of the final digital-to-analog conversion deep within a DAC is a very delicate analog process and is perturbed by minute energy levels at all frequencies. Since DAC designers (for the most part) only engineer to a specification, they perhaps don't regard such low energy levels as having an audible impact. So although the general effects of noise may be mitigated within a DAC, a proper attention to their effects is not fully considered.
So, yes, we want DACs to measure well and we want them to be well designed against the effects of noise - but we also know that attention to external factors helps. Isolation strategies such as using optical signals or low-noise power is desirable as are upstream digital components with a reduced RF footprint. How can we know which approach (or combination) improves the sound the most when the measured in-band noise is invariant?
Out-of-band energies are markers for what is going on inside a DAC and can quantitatively improve upon ad-hoc testing methods to determine how to best isolate a DAC from the effects of RF noise. Out-of-band measurements provide evidence that RF noise from the signal inputs, internally generated by the FPGA logic, injected by the power input or radiated from other components ends up at the DAC outputs. This means that those energies traversed paths within the DAC and my assertion is that there is a corollary between sound-quality and 'out-of-band' RF noise to strongly suggest a causal link between the two.
The measurement plots below were performed in a large RF isolation chamber (Faraday cage) using a Signal Hound BB60C Spectrum Analyzer and my own customized data processing and graphics output app. Signal tap was from the Hugo2's RCA (left) output through a ThorLabs EF500 DC Blocker.
Consider the measurement plot above. This is what a Hugo2's out-of-band signal looks like. The Hugo2 is powered up via its internal battery and in a quiescent state (no input) and just acting as a RF generator and antenna, frankly.
Now consider the addition of a MScaler that is also just powered up in a quiescent state (no input to upsample and no output). In this case, the MScaler is about 30cm (about 12") away from Hugo2. Notice the change in the RF energy measured.
The final plot shows the net contribution of an MScaler to the Hugo2's out-of-band signal (with energy differences subject to signal sweep and noise floor). Remember, this is all generated by RF noise radiated across open air - there is no signal or power connection between the two.
In rapid succession, I will be posting such plots for a number of typical upstream digital components and power supplies. I'll work to ensure the results are consistent wrt measured energy. I hope to generate a table of measured out-of-band net energy vs subjective sound quality. Stay tuned.