I’ve always considered myself an objectivist but lately, I’m less inclined to identify as that due to what I refer to as rabid objectivism. That doesn’t mean I am suddenly going to believe in magical rainbow stickers or purchase Headpie’s renowned audiophile bathwater, but at the same time, it does mean that I find myself distancing from individuals I once thought of as colleagues and friends. So what has pushed me to this point?
I wholly admit that measurements are important, but I refuse to believe they are the only things that are important, and failing to understand how those measurements are accurately and appropriately used is as big a fallacy as deeming measurements unimportant.
Lately there have been several examples of rabid objectivism in the forums that illustrate my point about misunderstanding measurements.
Balanced vs differential – I’ve written on this before balanced simply means the signal has an equal path to ground so a device that uses separate amplifiers and grounds for each channel is balanced. The argument lately that products were “fake balanced” because they have the same power as the single ended output is incorrect. The only difference in balanced and single ended is separate grounds vs a shared ground. In single-ended that shared ground means that neither channel has an equal path to ground as they have to contend with the return from the other channel. What these people are mistakenly causing balanced is actually a differential device that is composed of 2 amplifiers per channel and uses a positive and negative lead rather than a positive and ground. Differential does double the output power as voltage swings are from positive to negative rather than positive to zero. The obvious drawback being building a device with four amplifiers (two positive and two negative signal amplifiers) is twice as expensive as a device that only uses two positive amplifiers and a ground return.
SNR measurements are also commonly misunderstood. SNR is a ratio of signal strength to noise so higher is generally better, but even that has its limits. With most DACs being 16-bit we hit the first roadblock to the higher is better argument. Remember also that a 24 or 32 bit DAC decoding a 16 bit input file still has that 16 bit limit. If a 16bit DAC is perfect, the SNR is 96db. In the real world where most DACs get more like 12-13 bits of range the SNR is rarely over 88db. Once noise is introduced to a system it isn’t removed by later devices in the path and noise introduced before the volume control is amplified as volume increases, so if the DAC has an SNR of 94db (a realistic max for a 16 bit DAC), then that noise will be in the output and since it is introduced prior to the volume control on the amplifier it will be made more noticeable when amplified. Noise introduced after the volume control is not subject to being increased as volume increases. Why then are people currently obsessed with amplifier SNR of 130 dB or more, but these same people are content with 16 bit source material and DACs?
THD+N is a similarly misunderstood measurement. Almost all analyzers do THD+N because it is much simpler than measuring and calculating THD. To calculate THD, each harmonic has to be measured independently and then summed up. The decision of how many harmonics to measure makes THDs from different sources not comparable even though they share the same name. THD+N instead inputs a single frequency signal and uses a notch filter so it measures the output for everything but the input frequency (In theory). In practice, there may be some noise in the notch filters range that is excluded from measurements and thus the THD+N measured may be slightly less than actual due to measurement limitations. Here lately vanishing THD+N in amplifiers has been a huge topic of discussion with numbers like <.00005 being bandied about. Problem here is again the overall THD+N doesn’t tell the whole story. Realistically, we need to know something about the components make up that total. If most of the measured output is a mains hum at 60Hz, it may well be audible and have nothing to do with harmonics. If most output is lower harmonics it may be completely masked and a device with a fairly high THD+N may not have any audible distortion. Meanwhile higher harmonics are not masked and may be easily audible, so a device with a small THD+N may still have audible distortion.
To put all this into perspective, a lot of research by a lot of folks smarter than I am gives us some realistic targets for SNR and THD. Meyer and Moran conducted a study that introduced an extra analog to digital to analog conversion at 16 bit/44.1kHz in the signal path between the DAC and amplifier. They then used highly trained musicians, sound engineers, and others whose livelihoods require above average listening skills. In no case was any test subject able to determine reliably when that extra stage was added. Their conclusion, in normal listening situations, an SNR of 96dB or higher is transparent. If a digital volume control was added after the DAC it was found that 110 dB of SNR was needed to maintain transparency. Based on this, any SNR or higher than 110dB can be thought of as audibly transparent in all but the most extreme situations and anything approaching normal listening levels.
Likewise for THD+N the research indicates that even order harmonic distortion is audible only when at least 1% with odd order being more easily detected and the limit of audibility at 0.1%. Even including mains noise and other sources, it is reasonable to assume anything below 0.05% is inaudible.
So do numbers matter? Sure, but do they tell the whole story? No. I’ve never heard of anyone predicting what a headphone or speaker will sound like from the specs. Conversely some will argue that any two well-designed amps or DACs with the same specs should be indistinguishable. That might be true if every frequency was measured and compared, but measurements are done at a single frequency and often as not the specs don’t even list enough details of the testing to be sure two devices were measured the same way. Even if the two devices were measured in the same lab, measuring a single point doesn’t account for transducers with varying impedance. Almost all transducers impedance varies with input frequency. Balanced armatures I have measured with a nominal impedance of 16ohms vary between 7ohms and 92ohms depending on the test conditions. Other transducers have similar changes in impedance with things like the 600Ohm AKG varying from 385Ohms to 692Ohms in measured tests. Those same two amplifiers that measured exactly the same may react entirely differently to those shifts and may indeed sound different as a result.
I still consider myself an objectivist but even more so a pragmatist. Numbers are important, but they don’t tell the whole story and there are realistic limits beyond which the numbers mean little. Once you reach the point where something is inaudible, something with better numbers cannot be more inaudible regardless of how much “better” it is.
