romaz
1000+ Head-Fier
- Joined
- Jan 19, 2013
- Posts
- 1,182
- Likes
- 1,267
I won't be going back to USB anytime soon.
A New Perspective
I found some rare free time today and it has been awhile and so here goes...
USB vs SPDIF vs I2S vs Audio over IP vs other...which is the best data transmission standard for audio? I have struggled with these types of questions for a long time and I have been fed numerous different answers from reputable sources. You have the likes of John Mingo of Baetis Audio who will tell you AES is wonderful while USB is horrible and then you have someone like Mark Jenkins of Antipodes who will tell you the opposite and that his implementation of USB is better than anyone else's SPDIF. Looking at DACs, Berkeley refuses to incorporate USB at all in their DACs while the Chord DAVE embraces it. MSB utilizes their proprietary I2S in their CD transports yet Vince Galbo recently told me he believes the future belongs to their network renderer (AOIP) and not to I2S.
I believe the correct answer to which is best is "it depends." There are many examples of USB done poorly and so I don't blame those that proclaim USB to be an obsolete interface that should be relegated to laser printers and not high-end audio but if you look at the threads where people are passionately against USB, take note that none of those people are DAVE owners. I won't argue with a DAC manufacturer who claims their AES or ethernet input is superior to their USB input. In fact, it would be foolish of me to buy that DAC and not heed the recommendations of the manufacturer based on any pre-conceived biases I might have because truth be told, whether it be USB, SPDIF, I2S, AOIP or anything else, an interface is only as good as how well it was implemented. In my own experience, USB done properly can be as good (and even superior) to the other methods of transmission.
The appeal of I2S is that it separates the clock and serial data signals resulting in lower potential jitter compared to something like USB or SPDIF. With the DAVE, USB is isochronous asynchronous meaning the FPGA supplies the timing to the source and so incoming USB data is reclocked by DAVE's low jitter master clock. As Rob has mentioned many times, the DAVE is immune to source jitter up to an unheard of 2us and so this advantage of lower jitter that I2S claims to possess is rendered moot by the DAVE. I2S has problems of its own. First, this is a standard for serial transmission and not a standard for transmitting data and so most companies that utilize I2S (MSB, PS Audio, Wyred 4 Sound, Audio-Gd) have their own standards that are often proprietary. For example, you cannot connect an MSB I2S transport to a PS Audio DAC. Another problem you will find with I2S, because most implementations are proprietary is that 3rd party I2S sources are almost impossible to find. Sonore used to make a very good AOIP I2S device called the Signature Series Rendu that worked with a few I2S DACs like the PS Audio. While highly regarded, Sonore couldn't find a market for it and so it was discontinued. It is unlikely you will find any company investing much resources into a high-quality I2S source if the market isn't there. Finally, I2S is bandwidth limited and maxes out at 192 kHz sampling in a best case scenario. It is difficult to consider this an interface of the future with this type of limitation. Again, even MSB doesn't consider it their interface of the future and they were once its biggest proponent.
SPDIF, whether it be XLR (AES/EBU), RCA/BNC or optical has the same bandwidth limitations as I2S and so once again, it is difficult to consider this an interface of the future when 24/192 is the best that most SPDIF implementations are capable of (DAVE is unique since it's digital coax can pass a 384kHz sampled signal). My opinion is that those that claim that SPDIF is superior haven't figured out how to properly implement USB.
As for AOIP, I believe its future is bright but not necessarily at the expense of USB as the two share more similarities than differences. In fact, both USB and ethernet utilize packet protocols (they transmit data as packets) and so they both share the same strengths and weaknesses of packet protocols. Both ethernet and USB have the advantage of superior resolution over any SPDIF or I2S implementation. USB 2.0 (480Mb/s) and Gigabit Ethernet (1000Mb/s) are the only audio transmission protocols that can pass a 32 bit signal with sampling in excess of 384 kHz. For non-Chord DACs that benefit from HQPlayer with oversampling, for example, you must use one of these interfaces. SPDIF and I2S just won't cut it and so to say these interfaces represent the future would be to ignore what many cutting edge audiophiles are doing today with oversampling. When done well, both ethernet and USB are excellent and even complementary. A good example of this is the Sonore microRendu, what I consider to be the finest digital endpoint you can buy today regardless of price and this device incorporates AOIP on the receiving end but also USB on the transmitting end.
What are the challenges that face the integrity of both USB and ethernet transmission? According to John Swenson, the electrical engineer who created the microRendu, it boils down to 3 things:
1) Timing
2) Leakage loops > ground noise
3) Signal Integrity (SI)
Timing - Before asynchronous USB, this was a disaster but just about all modern USB DACs today incorporate asynchronous USB. If so, then why are some still complaining of problems with the SQ of USB? I think this is because not all asynchronous USB implementations are equal. Just because you remove dependency on the source clock and place it on the DAC's internal clock doesn't guarantee good results. You will see the likes of MSB and dCS charge upwards of $15k for their femto clock upgrades for a reason. Rob has already talked about his Pulse Array's constant switching scheme being innately immune to jitter without the need for expensive atomic clocks:
http://www.head-fi.org/t/766517/chord-electronics-dave/375#post_11963747
My own experience has convinced me that jitter is NOT an issue with the DAVE.
Leakage loops > Ground Noise - This is a bigger deal for DACs that have no galvanic isolation or poor implementations of it. Even though many DACs claim to be galvanically isolated, it is clear that galvanic isolation, which is defined as "resistance to DC" is not equally well implemented among DACs and clearly some DACs are better isolated than others. While Rob has admitted that his USB's galvanic isolation is not 100 percent, having recently tried an Intona Industrial USB galvanic isolator with the DAVE, I noticed no further improvement whatsoever. Further down in this post, I will reveal my experience with a variety of power supplies with my microRendu that has completely convinced me of the quality of DAVE's galvanic isolation to the extent that I now consider this a non-issue. While 100% of RF is eliminated with optical, as Rob has indicated, because his SPDIF inputs have to go through a buffer and then a DPLL before it is synchronized to the FPGA clock, USB is DAVE's best input.
Signal Integrity (SI) - The importance of SI is something that is relatively new to me but I am quickly understanding that this is the reason why some sources sound better with the DAVE. When I initially bought my DAVE, as I compared various sources that I had on hand, including several custom-built CAPS-type machines, a CAD CAT, and a variety of Aurenders, they sounded not significantly different than my basic Mac Pro or Windows laptop. Where the differences among these machines were much greater with my TotalDac, Bricasti and other DACs, with the DAVE, I found the differences to be small at best where it wasn't worth spending "thousands more" for one server over another. That opinion drastically changed with the Sonore microRendu. Even with a basic iFi switching PSU, while running Roon in bit-perfect mode, the difference against my Mac Pro and Windows laptop running on batteries was significant and I believe there are other DAVE owners who have shared similar experiences with their microRendu. Of course, this got me to wonder why this might be the case if DAVE is immune to jitter and is galvanically isolated.
As I started to follow John Swenson's posts and have asked him questions on the CA forum, I was intrigued by some of the things he said. This is what he had to say about why DACs with galvanic isolation and superior reclocking mechanisms are still susceptible to SI:
"What I have been finding in looking at DACs etc with USB inputs is that there is what I am calling "packet noise". This is bursts of noise caused by the USB receiver processing the packets of data. This noise shows up on both power and ground planes. Since the rate of packets is 8KHz there are strong components of this noise in the audio band. This noise can cause jitter in clock oscillators, reclocking flops, and DAC chips. It can also go directly into noise on the output of DAC chips.
The question everybody asks then is: well what about the DACs that have full isolation between the USB system and reclocking on the DAC side?Unfortunately this noise likes to make it through even this. Exactly how this works is complicated, I have written about this in the AudioStream articles. And bits and pieces in other posts recently. The upshot is that neither galvanic isolation nor reclocking completely get rid of it. They help attenuate it some, but don't get rid of it.
This packet noise consists of two parts: noise from the USB protocol engine and from the USB PHY. The protocol engine noise does not depend on the input signal quality, just the data, so its impact is always going to be the same no matter what is done with the input. The PHY is the part that actually connects to the electrical signals on the bus, ITS contribution to packet noise IS dependent on the quality of the input signal."
Exactly what is SI?
"SI consists of the rise/fall times of the signal, noise on the signal and jitter of the edges. Increases in any or all of these can decrease the SI. The decrease in SI can be so large that it becomes difficult for the PHY to determine the actual bits. Thus the PHY contains several methods used to pre-process the analog signals in order to make it easier to determine the bits. Modern high speed serial interfaces work at all because of these techniques that have been developed over the years.
When the SI is very good, the PHY can turn off the pre-processing steps and easily determine the bits. As the SI degrades the PHY turns on different parts of the pre-processing as needed. Each of these steps takes a fair amount of power to operate, thus creating noise on the power and ground planes. The more processing the PHY needs to use to determine the bits, the more noise is generated. Thus part of the packet noise is directly related to the signal integrity of the incoming signal. The higher the SI, the lower the noise.
It is very important here to realize this is noise that is GENERATED inside the DAC by its own operation, it is NOT noise on the USB bus that is somehow getting into the DAC as is commonly thought."
I came to another revelation when I finally found the opportunity to test various power supplies with the microRendu. John Swenson had spoken frequently of how a good PSU would make a very large difference with the performance of the microRendu. He discussed three properties of PSUs that he felt were very important.
(1) The first concept was leakage current which I mentioned above. Basically, any PSU, regardless of how well its designed and whether it is a linear design or not, if it is connected to the mains line, it will have a leakage current and this is what John had to say about it:
"...in a nutshell it is noise that flows in a loop through the mains, power supply, DC from the PS, through interconnects (analog cables, USB cables etc) to another box and then back through the PS of THAT box to the mains. This is NOT, I repeat NOT the "noise on the AC line" that everybody talks about, it is completely different, it is hard to grasp what it is and how it works, but it is there with ALL power supplies that connect to the AC mains. Different supplies have different amount and spectrum of this noise and every system is going to be different as to where this current flows. It is usually a fairly low level effect, but once most of the other issues are dealt with it is frequently The largest contributor to sound degradation.
Thus differing supplies can have different leakage current affects on a system, completely irrespective to anything having to do with the first part of this post. Thus you may find out that a supply with higher noise but lower leakage, can sound better than the lower noise higher leakage supply.
One major problem with specifying leakage current is that it forms loops, (which is why it is called a ground loop) it takes at least two supplies connected to the mains to form this, thus it is impossible to specify the leakage current of a specific supply, it HAS to be in relationship with another supply. That makes defining a number for a specific supply extremely difficult. (If you have more than two AC supplies it gets even more complicated)."
He said the only way to avoid or break this ground loop is to power either the microRendu or the DAC via batteries (meaning that at least one or the other or both has to not be connected to ground) or if there was complete galvanic isolation between microRendu and DAC. It was under this premise that he developed the LPS-1 power supply based on super-capacitors (like the Hugo TT). It incorporates two banks of supercaps and while one bank is discharging while powering the microRendu, the other bank is being charged by a feeder supply and switching from one bank to the other occurs automatically and seamlessly with the assistance of an FPGA. The importance of this design is that while a feeder PSU is plugged into the mains line to charge the banks of supercaps, the supercaps are completely decoupled from the mains line. In other words, the LPS-1 is completely devoid of leakage current.
(2) The second concept was output impedance of the PSU. I took this concept to heart immediately as it made sense to me. This relates to the ability of a PSU to provide instantaneous current when called for by the microRendu. This is apparently especially important with explosive transients in a musical passage. Here is what else John had to say about this:
"A bunch of experiments I did over a period of many years show correlation between signal integrity (SI) of the USB signal feeding the DAC and perceived sound quality. As the SI improves the signal sounds more "real". This has not been universally true for all people and all DACs, but has been true in a large percentage of the space (DACs and people).
Thus the purpose of the microRendu is to produce the highest USB SI I know how to do. SI consists of several things, jitter, noise on the signal,well formed signal (lack of over shoots ringing etc) and proper rise and fall times.
The power feeding the electronic circuitry is a very important part of achieving that high SI. As Barrows mentioned the noise on the power rails can cause increased jitter from the oscillator, and can directly wind up on the output signal.
The noise on the power rails comes in two parts: the inherent noise of the last regulator, (that is the noise when powering a fixed resistor) and the noise generated by fluctuating load current. This latter is particularly important for digital devices because the current drawn by the load (the digital circuitry) varies radically all the time. How the power supply behaves when given these changing load currents is very important. This is measured by an impedance VS frequency plot.
The Power Delivery Network (PDN) of the microRendu has been optimized to have very low impedance for high frequencies and mid frequencies, but not low frequencies. This was a deliberate engineering tradeoff, an important factor in USB SI is the length of the cable, shorter cables (everything else being equal) have better SI. So the device was designed to be very small so it could fit behind a DAC and connect with a very short cable. The tradeoff was that there is nowhere near enough room in such a package for enough capacitance to give goodlow frequency PDN impedance.
Thus for best performance of the microRendu the external supply must have very good low frequency impedance. Higher frequency impedance and inherent noise also make a difference but not as much as the low frequency impedance."
As I started asking around for the output impedance of linear PSUs that others were selling, no one could tell me except for Paul Hynes. Most suggested it isn't an important spec and others told me they didn't have the equipment to measure it. If you don't have the equipment to make measurements, I believe it is really difficult to know how good your product is. John Swenson obviously knows the importance of output impedance but because he indicated his scope isn't sensitive enough, he is incapable of providing these measurements at this time. He also indicated how important it is to know the output impedance of a PSU over a broad range of frequencies since impedance will vary with frequency. What are the measured output impedance values quoted by Paul Hynes: <3 milliohms from DC to 100kHz.
(3) The third concept was noise (ripple) and this is what most of us think about when we think about a superior PSU. This is also where linear PSUs generally excel over the switching types. Most switching PSUs have noise levels well in excess of 10 mV. The HD Plex LPSU that I own measures down to about 500uV, at least 20x quieter than the best switching PSUs. Both the LPS-1 and the Paul Hynes PSU have much quieter voltage regulators and so they quote noise levels as low as about 5uV. In fact, there are several very fine LPSUs that I found that have noise levels this low.
So this is where things became educational for me. When John first educated me about the concept of leakage current, because the microRendu requires only 6-9V at 2A, even before my LPS-1 arrived, I was able to test the impact of this leakage current in my own system. I compared the iFi and HDPlex PSUs against an inexpensive 9V battery supply I found on Amazon that is designed to charge your cell phone. It utilizes a noisy linear regulator (much noiser than either the HDPlex or iFi, I'm sure) but at least it wasn't connected to ground and so I could in some way directly assess the importance of leakage current. I did, in fact, notice a small improvement, enough to switch to this battery supply full time, but the improvement was quite small. I could have gone either way and it wouldn't have really mattered.
When my LPS-1 finally arrived, I believe I was the first to receive mine because of my proximity to UpTone Audio in central California. UpTone Audio manufactures the LPS-1 and so the LPS-1 is a collaborative effort between Alex Crespi (UpTone's owner) and John Swenson, the EE who designed the LPS-1 and also designed the microRendu for Sonore (a company unrelated to UpTone). Anyway, the LPS-1 almost doubled the performance of my LPS-1 against any of the other PSUs I had previously and so its impact was NOT subtle! This improvement obviously was not due to leakage current, or lack thereof, because using my inexpensive 9V battery supply, I had already determined that leakage current was not that big of a deal in my system (which I attributed to DAVE's well-implemented galvanic isolation) and moreover, the LPS-1 trounced this 9V battery supply. It had to be due either to lower noise (remember, this is about 5uV vs the HDPlex at about 500uV) or low output impedance. The improvement I was hearing, however, had much less to do with black backgrounds and more to do with increased dynamic energy (better bass to the extent that I had to turn down my subwoofer, better definition, better dynamic contrast, more realistic soundstage). If there was anyone that ever doubted that different bit-perfect sources can sound dramatically different with the DAVE, all they have to do is hear this combination and they would doubt no more and as far as I can tell, it has nothing to do with jitter or ground noise and more to do with the superior output impedance of this PSU driving the microRendu. For those interested in my more in-depth summary of the LPS-1 on CA, here it is:
http://www.computeraudiophile.com/f27-uptone-audio-sponsored/ultracap%99-linear-power-supply-1-launch-thread-29974/index8.html#post591892
By this time, my Paul Hynes SR7 power supply still had not arrived. Paul had been suffering through some personal problems and so it took months for this PSU to arrive but in my mind, I was almost embarrassed I had purchased the Paul Hynes supply as it cost more than 5x the LPS-1 and I thought that surely, this thing couldn't be better than the LPS-1. John Swenson was very clear that supercaps have the capacity to discharge extremely rapidly and so inherently, they have very low output impedance. In the defense of the Paul Hynes SR7, however, this PSU is a dual transformer, dual rail design (12V/5A and 7V/2A) and so it would have the capability of also powering a small NUC or Mac Mini or cable modem and so in my mind, I would be able to use it for other things. When I ordered it, Paul was given only one directive and that was to build the best PSU he knew how and so the SR7 is nowhere in the same league as his entry level SR3 or even his much higher performance SR5. Furthermore, part of the increased cost went into upgrades including silver DC cabling from the transformers to the DC outputs. DC umbilical cabling used was also fine annealed silver. Silver was specifically chosen for its superior conductivity (and therefore lower resistance) relative to copper. Ultimately, this should result in even better output impedance. I even went so far to specify his more expensive Mil spec XL connectors (which connect the DC umbilicals to his PSU chassis) as they have a very low output impedance of 1 milliohm. As I said, I took this concept of low output impedance very seriously as I was convinced this would be what would make the most difference.
So how did the Paul Hynes compare to the LPS-1? This, too, was very educational. To cut to the chase, the Paul Hynes improves upon the LPS-1. It is simply better and this was no small feat. While the difference between the Paul Hynes SR7 and LPS-1 was not as great as the difference between the LPS-1 and the iFi, HDPlex or cheap 9V battery supply, the difference was NOT subtle, nevertheless. In all the ways that the LPS-1 was better than my former supplies, the Paul Hynes was better than the LPS-1 in the same ways. Better dynamic contrasts, better bass, better definition, better soundstage, and obviously, in this case, it clearly had nothing to do with leakage current or mains noise since my Paul Hynes SR7 was connected to ground and the LPS1 was not. Because both the SR7 and LPS-1 had similar ripple noise values, the difference I was hearing most definitely had to do with output impedance (in my mind, at least). This also convincingly proved to me that battery supplies are NOT automatically better.
This is not where my education concluded, however. There was much more to discover. Remember, this Paul Hynes SR7 is a dual rail design and a 12V rail can come in handy for many things. When I ordered this second rail, I was intrigued by another statement that John Swenson had made. In essence, what he said was that most devices he knew of benefited from better SI. The contradiction of this statement, however, is that he claimed the microRendu, like the DAVE, is largely resistant to what comes before it. If you recall, the microRendu accepts an ethernet input and ethernet is inherently galvanically isolated. If ethernet has one natural advantage over USB, this is it. On top of that, John had built into the microRendu ethernet filtering and so he was not detecting much difference among various ethernet cables or whatever digital source he was using to feed the microRendu (ie Mac, PC, laptop, sonicTransporter). In my own testing, I was able to verify that indeed, there wasn't any significant difference among sources that I owned but then again, none of the sources I owned were being powered by a superior PSU with ultra low output impedance.
I dug up my old NUC Windows PC that I had long ago shelved because I preferred the convenience of my Mac Pro. The benefit of this small computer is that it has a 12V DC input and so I went ahead and powered it with the 12V lead from my Paul Hynes SR7. Without any special OS optimizations, I installed Roon Server and compared it against my Mac Pro feeding my microRendu. To my amazement, there was improvement! The improvement was not as great as the difference between the Paul Hynes SR7 and the LPS-1 but once again, the difference was not subtle and was easily discerned to the extent that blind testing was never considered necessary.
But wait (you guessed it), there's more. If improved SI to the microRendu from the Roon server (NUC) that fed it resulted in noticeable improvement in SQ, what would happen if I improved the SI to the NUC by powering the cable modem streaming Tidal to the NUC? With my previous setup, local storage and Tidal streaming were always very close but local storage (especially with my very best CD rips and downloads) was often a bit better. Well, I went ahead and powered my TP Link cable modem with the 12V lead from my Paul Hynes SR7. The cable modem was then connected via CAT6 to my Netgear WiFI router which was then used to connect directly to my NAS (local storage) and my NUC. I listened to a variety of my best sounding 16/44 files that were stored on my NAS and compared them to the equivalent 16/44 streams from Tidal. These were files that I had established beforehand sounded a bit better (smoother) compared against their equivalent Tidal streams. Now, with the Paul Hynes SR7 powering my cable modem, not only were the Tidal streams as good as streaming from my NAS, they were actually a little better and more dynamic. While the difference was not as great compared to the SR7 powering the NUC, the difference, nevertheless, was NOT subtle! Because my NAS takes more than a 12V input, I was not able to test what would happen if I powered my NAS with a superior PSU but it has left me wondering.
Are all improvements upstream noticeable? I have not found this to be the case. I have been playing around with different ethernet cables (AudioQuest Diamond, SOTM CAT6, Supra CAT8, Blue Jeans CAT6A, optical ethernet with TP Link FMCs) and the differences here have been small. I even purchased a Paul Pang ethernet switch with TCXO clock (an "ethernet Regen," if you will) to reclock the signal just before it arrives to the microRendu and powered it with my LPS-1 thinking that this could be a way to improve SI further to the microRendu and while there has been an improvement, it is very minor at best. Thus far, the biggest bang has been to power a device with my SR7.
Equally importantly, if I switch back to the iFi as a power source for the microRendu, any improvements made upstream are much much less difficult to appreciate and so powering the microRendu with the SR7 is where the greatest difference is to be had in my system.
Of course, this has made me more than a little curious about what exactly makes a Paul Hynes PSU so special and so I asked Paul. In his opinion, output impedance is not the only important parameter that he looks at and here is what he had to say:
"Power supply output impedance is an important parameter, as are transient response, settling time, operating bandwidth and noise. Neglect one or more of these parameters in the design stage and you will have noticeably lower performance. For exceptional power supply design it is important to consider all of these parameters and optimise them to the best of your ability. This is what I do. The ideal power supply would have zero impedance at all frequencies of operation as you cannot develop any voltage into zero impedance no matter how much current passes through it. In the real world all circuits have some level of impedance and any current passing through this impedance will generate a voltage fluctuation, which in reality becomes an additional noise source in the system, which degrades signal integrity. It is therefore important to minimise the impedance to reduce this disturbance to the lowest level. This impedance reduction is usually achieved by an error amplifier using high levels of negative feedback, which introduces all manner of problems with operating bandwidth, transient response and settling time. This is a big subject to consider so I will not go into detail here. Suffice to say I do not use typical circuit topologies in my voltage regulator designs to achieve low impedance over a very wide operating bandwidth."
As to how the signal so high up in the chain (from the cable modem to the NUC to the microRendu) can ultimately impact something as good as the DAVE that is immune to source jitter and has superior galvanic isolation, here is what Paul had to say:
"All circuits require a reference to operate and react with each other. This is typically via a ground (0V) system. Some circuit sections also require a voltage reference above (or below) ground to provide precise operation and this voltage reference is usually connected to ground for its own reference. If the power supply is not clean and free from noise and transient disturbances, it will pollute the ground reference and anything connected to it creating uncertainy of reference voltage. This noise and transient disturbance can be passed on from stage to stage once embedded in the signal and it is quite capable of causing timing erros due to data streams. Reducing the magnitude reduces uncertainty."
Essentially, John Swenson has said the same thing, that there are anomolies that can occur upstream of the DAC largely due to issues in the power distrubtion network of various digital components that have the potential to "embed" themselves into the output signal to the extent that once this compromised signal reaches the DAC, there is no way for the DAC to know how the signal has been compromised and is defenseless against it. Whether this is true or not, I will leave it to each of you to make up your own minds but my observations (which I have confirmed several times now and in the company of others) suggest to me that they are.
In summary, in my opinion, the DAVE is as good as Rob has said it is. Its galvanic isolation is well implemented and it is immune to source jitter. I do not believe my observations above have anything to do with the competence of the DAVE nor do I believe that they can be salvaged by any DAC. I have yet to hear a source sound badly with the DAVE but the quality of upstream components, especially the quality of the power supply network to these components do absolutely matter and can result in stunning levels of improvement with any DAC but I believe the improvements are probably best realized by the best DACs. While there is no substitute for a good recording, there is so much more in all of our recordings that we are not hearing because of these anomalies that up until now, have not really been considered. While I'm sure discussions about USB vs AOIN vs I2S vs SPDIF will continue, I believe there are perhaps bigger issues that will now need to be addressed first.