The science of high-end audio has to go beyond accepted science for a variety of good reasons. In my experience, many scientific fields that are trying to optimise performance, as opposed to just make something work, become at least as much art, as science. But the most fundamental reason is that accepted science works to strict rules that get in the way.
For very good reasons, guesses become accepted science through testing in narrowly controlled experiments that can be objectively and repeatedly measured to yield the same result. This draws the conventional science of audio away from the true objective, to focus instead on the science of the equipment. The equipment can be measured objectively and consistently. But the true objective is to eliminate whatever distortions get in the way of a listener getting the same emotional benefits from reproduced music as if the listener was hearing the actual performance. This involves the science of the ear-brain system, but it is not as easy to measure as the equipment is.
This argument goes beyond the notion of psycho-acoustics or placebo. The ear-brain system, that interprets changes in sound pressure at the ears into recognition of separate instruments and voices, each in their distinct spaces, is not a passive system but a very complex active process that we know very little about. To ignore the ear-brain process, and not seek to understand what forms of distortion have the worst impacts on our enjoyment of music, is only valid if the science of the equipment can lead to perfect equipment. But how would we ever know our equipment is perfect if we are never measuring what is relevant?
The problem of course is that human responses to music are not considered objective measurement and are rarely able to give you consistent results. As Einstein is quoted to have said “not everything that matters can be measured”.
For this reason high end audio firms have to rely on their experience as music lovers to push beyond the limited performance achieved by only focusing on the science of the equipment.
In digital audio, so much of the conventional wisdom appears to me to be 100% wrong. And yet many commentators are utterly adamant about what can and cannot matter in digital audio. This delusion that a scientific theory is somehow an immutable fact or law is entirely unscientific. Real scientists appreciate that even our best theories will only ever be our latest best guess at how reality works. Perhaps more importantly, even if one day our theories do completely describe reality we will not know for certain that this is the case.
When someone insists on an immutable scientific ‘fact’, he or she is stating their faith and are not representing true science. Just as importantly, the opinions I express below are just my humble insights collected over the years in case you are interested. I do not intend to state them as facts and I may very well have experiences in the future that change my mind on any number of these issues. If you came here hoping to find scientific facts about computer audio you will have to seek that delusion elsewhere.
Digital audio science suggests that any old server will do provided the correct bits get to the DAC. Even if there is jitter or noise in the digital signal, if the bits can still be read then the signal can be regenerated and reclocked so that the DAC chip receives as accurate a digital signal as it is possible to provide.
But listening in the real world suggests that this is simply not true, and misses something. This theory is just an assumption, deduced from mental models of digital signal processing and unsupported by relevant experimentation. In fact experimentation suggests to us that the complete reverse is true. Our experiments convince us that if you clean up the digital signal, or never pollute it in the first place, the resultant audio improves regardless of what buffering, re-clocking or regeneration occurs along the way.
DAC manufacturers have repeatedly claimed that you can use any old server, and their DAC will clean up the signal. But in our experience the reverse is equally true. With a poor server, then better DACs can make a significant improvement. But if you improve the server, and therefore improve the digital signal sent to the DAC, not only is the sound significantly better, but the differences between DACs is diminished. We might flippantly state that with our servers any old DAC will do, but we are not so biased as to claim that.
The same false deductive logic has in recent years suggested that you can use any old device to run the server app and you just need a well-designed renderer device and get great sound. Again, this is simply not supported when you listen to the impact an excellent device running the server app can have.
It seems that the false deductions are all based on the mental model that we are only dealing with 1s and 0s. This is a false understanding of the task. There are no 1s and 0s. They are a logical construct. An abstraction from reality. The reality is an imperfect pit in a disk, a light turning on and off in the presence of reflective surfaces, an electrical square wave signal that cannot perfectly square out the wave (since it does not have infinite bandwidth) and that is polluted with noise interference. Once you appreciate this, then you may begin to appreciate why making the representation of 1s and 0s easier to discern at every step of the process makes the result more accurate downstream at the final DAC step. Or you could listen to some good music server products, and then you will know.
Antipodes Audio music servers are designed from the ground up to minimise noise interference with the digital audio signal. In our experience, success at this allows the beauty in your music be heard.
In simplistic terms, the most challenging problem that has to be dealt with in designing any audio component is noise. This is most true in digital audio, which may seem to be a paradox since digital audio is meant to eliminate background noise, and those 1s and 0s are meant to be immune to it.
Noise interference with a digital audio signal does not result in the listener hearing noise. It results instead in the timbre of voices and instruments sounding unnatural, and spatial clues becoming confused, and we have all heard this with poor digital.
It is not challenging to communicate the bit data accurately to the DAC. But any distortion of the digital square wave has the effect of making the DAC's job harder to do, because it has to do more work to decipher the received signal. It is important to realise that if you increase the amount of work a chip has to do then its accuracy diminishes. A fluctuating workload also has the effect of fluctuating current requirements which impact on the stability of the power supply applied to other parts of the process.
Whatever the process issues are, our conclusions are that if you reduce noise interference with the digital signal, it has a significant effect on the sound coming out of any DAC. And conventional computers are very noisy electrically.
So, in simple terms, Antipodes Audio builds servers from the ground up to minimise noise. But it is not as simple as just reducing noise. Noise is not one dimensional, but has both an amplitude and frequency component. Reducing the amplitude of the noise is beneficial, but changing the frequency of the noise also has a big impact.
One way that the frequency spectrum of the noise is important is that low frequency noise does less damage to a digital signal than high frequency noise. So a range of design choices are made throughout Antipodes music servers that shift noise down in the frequency range.
Another way that the frequency spectrum of the noise is important is this. Each active component in a circuit will generate a certain spectrum of noise. If two components create noise at the same frequency, or overlapping frequencies, then the noise level adds together and the noise floor increases. Therefore design choices are made throughout Antipodes music servers, such as in parts selection, and to alter the clock rate of certain chips, to shift the frequency spectra of the noise and thereby minimise noise "nodes".
Finally, the production of a digital square wave requires very high bandwidth and phase accuracy. Noise filtering a digital signal will damage bandwidth and phase. Some music servers use noisy components like switch mode power supplies and spinning hard drives and then add noise filtering to address the problems these parts will inevitably create. But using noise filters limits bandwidth and slows impulse response. Therefore Antipodes Audio goes much further in ensuring its servers do not create noise in the first place, thereby eliminating the need to introduce noise filters. No bandwidth-limiting or phase-shifting noise filtering is used, and we have found that the better we do this, the more the music has natural speed, immediacy and dynamics - something that our servers have become noted for.
Regardless of the science of how we do it, all of our endeavour is simply about getting you closer to the beauty that was in the music you love when it was made. You need to be the judge of whether it is worth it for you. If you are a music lover, we believe it will be.