Frequently Asked Questions
Some people insist that any computer will do for the music server/player job because the data is just 1s and 0s. Why do I need a high-end music server/player? ANSWER
How would you describe Antipodes Audio’s design philosophy? ANSWER
What is the best connection to use between a music server/player and a DAC? ANSWER
Do all Antipodes Music Server/Players perform the same functions? ANSWER
High-End Music Server/Player
A music signal is two-dimensional – it has an amplitude dimension and a time dimension. We only hear sound when the amplitude oscillates over time. The greater the amplitude, the louder the sound. The faster the amplitude oscillates, the higher the pitch of the sound. A sound system should aim to accurately reproduce both dimensions.
Encoding an analog signal as a digital signal involves sampling amplitude measurements at regular time intervals (at the sample rate). While the black digital ‘staircase’ line below looks very different from the red actual signal line, the DAC stage converts the data back to a smooth wave.
The point to understand is that the data in the digital music file represents only the amplitude measurements, and there is no equivalent recording of timing data because the file simply states the sample rate in a header, so that the playback process knows the rate to play it at.
Because the bitrate is finite, the digital audio recording process will always require trade-offs that result in small errors. But this is, arguably, offset by the resilience of a digital file when storing or transmitting it, when compared to an analog signal which inevitably and irrecoverably degrades with every processing or transmission step.
Binary data (using multiple 1s and 0s to record each amplitude measurement) is easily transmitted from one point to another without distorting the information. The binary 1s and 0s can be represented using high (for a 1) and low (for a 0) levels of voltage or light (or even holes in a punch card or different magnetic charges, etc).
The key point is that the voltage level or light intensity does not have to be perfectly accurate. Mild to moderate levels of distortion will not cause a high value (1) to be misread as a low value (0), or vice versa. In the image below, any signal over the high threshold is read as a 1, and any signal below the low threshold is read as a 0. The image below illustrates how it takes extreme signal distortion to create a bit error.
High-bandwidth data networks can easily move digital data files quickly over long distances without any distortion of the information contained in them, provided the network is designed to keep the signal level above the noise level, and to simply re-send the occasional packet that arrives corrupted at the other end. This works very well for transmitting web pages, emails, spreadsheets, etc. It also works for transferring video and music files. But the situation is different when you are streaming video and music files, because timing becomes a factor.
IT IS ABOUT TIME
The digital audio signal that enters the DAC chip needs to have accurate and unambiguous timing, or the resulting analog audio signal will be distorted. Ideally, it would be a perfect square wave, as in the blue line below. If the timing is distorted then the resultant analog audio will also be distorted.
This means that playing a music file is a different problem from just transmitting a file from one place to another. Accurately playing a digital music file means playing the amplitude dimension accurately, with accurate timing. The key point here is that the Achilles heel of digital audio is that sound quality is hyper-sensitive to the timing errors that are inevitable in the transmission of digital data.
The challenge for a Music Server/Player is to send a perfectly-timed square wave signal of the music’s digital file. But it is impossible to produce a perfect square wave because of the real-world presence of bandwidth limitations and noise interference. A perfect square wave would have infinite bandwidth so that the signal could be at its lowest value and at its highest value in the same instant when transitioning between a 0 and a 1.
In reality, bandwidth is limited and so there will be a rise time and a fall time.
There will also be some noise interference with the signal from a variety of sources, as always occurs in any electronic environment. Noise above the bitrate adds the wiggles shown in the graph below.
Noise below the bitrate will make the whole wave rise and fall.
The combined effect is illustrated below, and shows how timing errors are inevitable.
The challenge is to maximise bandwidth and minimise noise in order to minimise the distortion of the audio signal.
In an analog environment, electronic noise adds to the signal but does not necessarily change the sounds of reproduced voices and instruments because the noise is often heard as a separate sound. In digital, the combination of noise and bandwidth limitations will always produce some timing error, and this is not heard as noise, but is heard as changes to the reproduced sound of voices and instruments.
When someone claims that you can use any computer to do the music server task because it is just about 1s and 0s, they are completely right about the amplitude dimension, and completely wrong about the time dimension. While the music file tells the Music Server/Player the rate at which to play the file, the timing accuracy achieved varies with the quality of the design and implementation of the Music Server/Player and DAC. While digital audio makes the amplitude information very resilient, it also makes the timing information hyper-sensitive and hyper-critical to achieving a high-end audio result.
If only moderate fidelity is required, then the playback system can be rudimentary. The sound will be fairly clear and articulate, but the sounds of voices and instruments will be unnatural, and listener fatigue will set in quickly.
Getting the timing right is not only very hard to do very well, but is impossible to do perfectly. Like everything else in audio, getting ever closer to perfection is a never-ending challenge. What is good enough for you is a personal choice, and not something that the science can determine for you.
People sometimes make the observation that despite its weaknesses analog audio degrades more ‘gracefully’ than digital. In the decades since the first terrible-sounding CDs and CD Players, digital audio recording engineers and digital audio equipment manufacturers have made some big improvements. This is not because digital audio theory has been changing. It is because we are learning and discovering how to reduce the problems in digital audio, and we are learning about making their impact on music more ‘graceful’.
There are several parallels in the audio industry. For example, in the early days of solid state amplification, the differences from the best tube amplifiers were stark, but over many years the sound of great amplifiers, tube and solid state, has converged. The same thing is happening in the source world. Digital sources and analog sources still sound quite different from each other, but the sound of the best examples continues to converge through advances in design.
The science of digital audio makes things sound simple. In practice, the processes require high levels of processing power and wide bandwidth, at the same time as keeping noise interference very low. These requirements work against each other. For example, higher power generates higher noise levels, and filtering noise limits bandwidth.
Antipodes minimises the trade-offs between clocking, noise and bandwidth by:
- designing its circuits from the bottom-up for low noise, which reduces the need for noise filters
- designing very fast low-noise power supplies to support the bandwidth requirements
- designing around optimisation of a four step process.
In our four-step approach (see the diagram below), the first two steps use asynchronous streaming, buffering and regeneration of the stream. Step 1 uses high processing power and Step 2 repeats the process with low processing power. The key objectives are to achieve low noise and high bandwidth. Because the first two steps use asynchronous transmission, a very high-precision clock at the sending end is not warranted. For example, in asynchronous transmission, the receiving electronics will command the sending electronics to slow down or speed up to make sure the buffer does not empty-out or over-flow.
The last two steps use synchronous signal transmission, re-clocking the signal with reference to an ultra-high-quality clock. Step 3 re-clocks using high processing power and Step 4 repeats the process with low processing power. The high-power Step 3 also has to convert the asynchronous signal to a synchronous signal. In the last two steps, transmission is synchronous and so the key objectives are precise signal clocking, low noise, and high bandwidth. This is where using an oven controlled circuit and ultra-low phase error clocks are crucial but, as is always the case in digital audio, the power supply has a huge influence on the final result.
Note that the fourth step occurs in the DAC where it is almost always a low-power processing step. Preceding it with a high-power re-clocking step will lift the performance of any DAC.
The need for precise clocking appears to be well recognised, and many discussions focus heavily on the clock used, but the effect of the clock on timing is only as good as the combined weaknesses of the clock, the re-clocking circuit, the circuit layout, the other electrical parts, and the power supply. The power supply has a huge impact on all digital circuits, because it is a major challenge for a power supply to deliver the required speed for high-power high-bandwidth processing, at the same time as keeping noise very low.
It is sometimes claimed by DAC manufacturers that their DAC completely eliminates timing errors. But, by way of example, an APLL (which is often used for re-clocking) is similar to a negative feedback system. The timing in the signal received is compared to a reference clock to determine the timing error in the signal. The error is inverted and added to the original signal in an attempt to correct the timing error. Just as using negative-feedback does not make amplifiers perfect, real-world challenges mean that APLLs, ADPLLs, etc, vary in quality and none ever deliver a perfect result. Some re-clocking techniques are much more sophisticated, but the point remains that perfection is never attainable. Feeding the DAC with a better signal from the Music Server/Player always improves the sound quality.
The four step process is designed to improve timing accuracy at each of the successive steps. As an analogy, imagine your car is caked with mud and grime. You might take four steps to get it very clean.
- Water-blast off the big pieces of dirt.
- Use a sponge and warm soapy water to clean it further.
- Rinse with fresh water and towel it dry.
- Apply wax, and polish.
If you skip any step, you compromise the end result. Similarly, if you improve the quality of any one of the steps, it is very likely that you will improve the end result. And this is true too of the four step digital audio playback process we have described above.
In digital audio, even more steps may improve the timing further. For example, there are several products that are claimed to improve timing at various points in the process, such as network switches, network bridges, signal regenerators, etc. We have found that the money is better spent on improving the quality of the four steps outlined above, rather than adding more steps. So we recommend you look to buy a better music server/player than buy a less capable one together with these add-on products. But you can use these add-on products to improve the performance of a music server you have already invested in, provided you select them well. Beware that in the wrong context, many of these add-on products can reduce performance.
Connecting To A DAC
We often see discussions about what the best connection is, between a music server/player and a DAC – Ethernet, USB, S/PDIF, AES3 or I2S. But this is not a question about different connection types. It is about different solution architectures. This is because Ethernet, USB and the synchronous interfaces (S/PDIF, AES3 and I2S) occur at quite different stages of a computer audio solution, and so are not direct alternatives.
The image below describes the four step approach outlined in the previous section on Design Philosophy. The purpose of the steps is to take a stream that has poor clock timing accuracy, or a file that has been transferred from a storage disk to RAM in block mode, and progressively create a precisely timed digital audio signal for the DAC stage. You can certainly make the music play with a simpler set of steps, but to get a high-end audio result, more steps are necessary. We recommend that you read the Design Philosophy section if you want to learn more about the process steps.
In the early years of computer audio, people used basic computers for Step One. This placed a heavy burden on DACs to do the rest, so DACs sprouted Asynchronous Ethernet and USB inputs. In the absence of quality music servers, DACs had to include more of the computer functions in order to improve signal timing before the DAC chip.
Over the years audiophile music servers emerged, they got progressively better at the job, and audiophiles began to realise that using a standard computer did not get the job done nearly as well as using a good music server to feed the DAC, regardless of the quality of the DAC.
For even better sound, the Antipodes Oladra and K50 models take the additional step of performing the asynchronous-to-synchronous high-power Re-clock stage before the DAC. A Re-clock stage performed outside the DAC can employ much higher processing power. Placing a high-power Re-clock stage inside the DAC would create high levels of noise interference. Without a separate high-power Re-clock stage, DAC design has to trade away processor power to keep noise down. By placing the first Re-clock stage in the music server we can give it all the power and parts quality required to do the job to the highest possible level of accuracy. Your DAC will also perform a low-power signal Re-clocking before the DAC chip, and feeding this with a high-precision signal from an Oladra or K50 will dramatically improve the sound quality achieved.
Some of the top DAC manufacturers clearly agree with us, producing multi-box solutions where the first Re-clock stage and sometimes also the Player stage are in a separate case or cases from the DAC stage. In this way, they can apply high processing power and avoid generating noise interference inside the DAC.
Of the synchronous connections, I2S is better than AES3, and AES3 is better than S/PDIF. The differences are not so large that S/PDIF, using a high-quality S/PDIF cable, cannot out-perform I2S using a basic cable, for example. But I2S has the advantage of being able to handle much higher bit-rate transmission, and a clear channel for transmitting the clock data.
An ideal cost-no-object design would, arguably, place each stage in its own separate case. When price is a consideration, this ideal can be traded-off by placing some stages together in the same case. Which ones the manufacturer puts together determines the type of connection you use between the music server and the DAC.
For example, with an Oladra or K50, we recommend you use a synchronous connection rather than USB or Ethernet. This is mainly because you get the added benefit of the high processing power Re-clock stage in the Oladra and K50. But you also avoid using the inherently noisy Ethernet and USB stages inside your DAC.
In the same way, a DAC manufacturer perceiving that most of its customers are still using basic computers as their servers, will tell their customers that their DAC’s Ethernet connection is best.
Which one of us is right depends entirely on the Music Server/Player and DAC that you use.
For the reasons outlined, the choice of connection between your music server and DAC is not about differences between the connection types. It is a decision about the ideal composition and architecture of your computer audio solution.
- If you want to keep the computer and higher-power stages (Steps 1, 2 & 3) away from your DAC, to avoid noise interference with the DAC and analog stages, then connect your music server/player to your DAC with a synchronous connection, S/PDIF, AES3 or I2S.
- If you want to only keep the Server and Player stages away from your DAC environment, and have the DAC perform the Re-clock and DAC stages, then connect your music server to your DAC with a USB cable.
- If you want the Music Server to only do the Server stage, and leave the rest to your DAC, then connect your Music Server to your DAC with an Ethernet cable.
If you have an Oladra or K50, then the first option above will give you the best sound quality. The other two options mean you can use a simpler music server product, and this may make sense depending on the design and capabilities of your DAC. There is only a small number of two-box DACs that can perform the Player and/or Re-clock stages with the power and isolation required to compete with the Oladra and K50 but, in these cases, using your music server to only perform Step One can make sense.
The diagram showing the four process steps is helpful for describing the differences between the models.
ANTIPODES DS, DV, DX – Medium-Power Music Server/Players
Up until around 2015, all Antipodes models were designed to complete Step 1 and Step 2 using a single computing device. This refers to the DS, DV and DX models. A single computer was used to run both the Server app (Step 1) and Player app (Step 2), and output via asynchronous USB to the DAC. The USB output on all Antipodes music server/players is 100% compatible with the USB Audio 2.0 standard. The main downside of this approach is that you need to decide on whether to design the computer to be high-power or low-power. High-power benefits the Server processes but creates too much noise for the Player processes. An additional problem is that the precision of the Player process is also undermined by the large number of system processes that need to be run to support both the Server process and the Player process. Another problem at the time was the poor design of USB inputs on many DACs.
ANTIPODES CORE, DS CORE, CX, S40, K40, K41 – High-Power Music Servers & Music Server/Players
These models were each designed to be used for just Step 1. They employ a single high-power computer optimised for running Server apps. The K40 and K41 do not have Player apps available to be used, and so their output is Direct Ethernet. Direct Ethernet means the Server device is connected to the network (for user control, internet streaming services, and network-distributed audio), and then the Server device is connected directly to a nearby Player device by an Ethernet cable.
But the CORE, DS CORE, CX and S40 can be used as a Music Server/Player, to perform Steps 1 & 2. The reason for this is to allow a customer to start with one of these models, and to add a low-power Antipodes Music Player later, to perform Step 2 independently. Therefore, these models have Server and Player apps installed, and both Ethernet and USB outputs.
ANTIPODES EDGE, DS-X, EX, S30, K21 – Low-Power Music Players & Music Server/Players
These models were each designed to be used for just Step 2. They employ a single low-power computer optimised for running Player apps. But they also have Server apps installed, enabling a customer to start with one of these models to perform both Steps 1 and 2, and to add a high-power Antipodes Music Server later, to perform Step 1 independently.
ANTIPODES S20 – High-Power Re-Clocker
The S20 is designed to perform Step 3. The S20 takes a USB feed from any of the above Antipodes models (except K40 and K41), galvanically isolates the USB from its re-clock stage, and provides precision-clocked synchronous outputs: S/PDIF, AES3 and I2S, to send to the DAC.
ANTIPODES K22 – Low-Power Music Player, Music Server/Player & High-Power Re-Clocker
The K22 is designed to perform Steps 2 and 3. It is equivalent to the K21, with the addition of an ultra-high-precision high-power re-clocker. The K22 can also run Server apps, and has Ethernet, USB, S/PDIF (TOS, RCA, BNC), AES3 (3-Pin XLR) and I2S (RJ45, HDMI) outputs. The K22 can be upgraded by adding a high-power Music Server later, to perform Step 1 independently.
ANTIPODES K50, Oladra – High-Power Music Server, Low-Power Music Player, High-Power Reclocker
The K50 and Oladra are designed to perform Steps 1, 2 and 3. Each of these models have a high-power computer dedicated to the Server apps, a low-power computer dedicated to the Player apps, and a high-power engine, featuring OXCO-clock, FPGA processing and Galvanic Isolation, to convert the incoming asynchronous signal from the Player to a very precisely timed synchronous signal to transmit to the DAC. The K50 and Oladra have Ethernet, Direct Ethernet, USB, S/PDIF (TOS, RCA, BNC), AES3 (3-Pin XLR) and I2S (RJ45, HDMI) outputs.
Word Clock Output
The S20, K22, K50 and Oladra offer a Word Clock Master output. This provides access to the two ultra-low phase error clocks used in the re-clocking process. One clock is dedicated to 22.05MHz for sample rate multiples of 44.1kHz. The other is dedicated to 24MHz for sample rate multiples of 48kHz. The clock output is designed for use with synchronous digital audio signals, and has not been designed for general purposes. There does seem to be a general lack of appreciation that different purposes require different clock chip designs, let alone different circuit designs. The word clock output on our products should only be used as a ultra-high precision Master clock reference to slave your DAC to, and is not suitable for other devices like network switches, etc.