The USB cable mega thread. Walking off the plank

From a conceptual point of view, digital is digital. If the receiver receives the data, it does't matter how the data got there. It is the same regardless. This is a fair argument. Unfortunately if you've ever been responsible for PHY design, particularly of high speed PHYs, you'll discover that life isn't that easy.

What matters at the receiver is signal integrity (and power integrity if that applies). When a modern PHY sees a data signal, it tries to receive/interpret that signal. If the signal is a mess, it starts using analog techniques to recover it, trying to make the data eye as pristine as possible and placing the clock in the best possible spot.

The good news is that these techniques usually allow for data recovery and your system is happily on its way.

The bad news is that we are talking about having critical, sensitive analog components after the receiver in these DACs.

This is where the PHY design is critical to the application.

Any time the PHY has to turn on features to recover a signal, it uses power. Increased power consumption by the PHY often makes the PHY more susceptible to noise. If the PHY isn't designed well ( or with this I'm mind) noise from the power supply or data can couple into adjacent power supplies and data lines. (imagine a guy yelling at you at your table in a restaurant. You can hear him and understand him, but his yelling affects what other people can hear at the tables around him).

What's easy to forget (particularly for digital logic designers) is that everything digital is really analog. Digital is simply an interpretation of the analog. And analog loves to pick up noise (which is why we buy these expensive things to begin with!)

Ok, so what about cables. Cables are transmission lines. They contribute to the signal (and power) integrity experienced by the receiving PHY.

The concepts behind usbRegen and Wryd (among others), is to make the receiving PHY see a nice clean signal(and power). Seeing a nice clean signal makes it such that the Receiving PHY doesn't have to work hard to resolve the data and reduces the likelihood of noise (be it power or data) seen by the PHY being coupled into the down stream work the DAC is doing. Retransmitters also fix the signal if the noise source is at the transmitting PHY. Something that can never be done with a passive cable.

Whether you are reducing noise by buying a better cable or by retransmitting it closer to the receiver, you are trying to accomplish the same thing.

As an analog engineer, I'll tell you that you are better off retransmitting pristinely very close to the DAC receiver. The cable before the retransmission won't matter much if the Wryd or regen is designed well. The cable after the retransmission should be as short as possible. The longer the cable, the higher quality cable you need.

If you don't have a retransmitter, then a better cable will reduce noise and improve signal integrity (to a lesser degree) if the transmitting PHY isn't the one spewing garbage (garbage in, garbage out).
Shorter cables are better. (Short cable after the retransmit is great).
Thicker gauge wire is better.
Fully shielded wires and connectors are better.
Pure copper is good.
Pure silver is expensive but good. Unlikely to produce enough benefit to provide good value.

Whether the improved signal integrity improves the sound depends on the USB implementation inside the DAC. Many companies use off the shelf USB receiver chips. Very few (none?) of these chips are designed to feed into sensitive analog components and as such, their designers didn't pay as close attention in the design to coupled noise as they would if they were designing specifically for these DAC applications. (They are making parts to sell in huge volume cheaply).


TLDR; The USB signal received by the DAC can affect the analog functionality inside the DAC. Nice cables are a way of reducing this. Retransmitters are another way. A retransmitter will reduce this more than high end cables. A well designed USB interface in a DAC can also reduce this.



Disclaimer: I don't work for any audio or electronic company that could possibly gain from this forum. I don't want to promote any particular product. I'm not particularly an audiophile (yet). Though I am an analog-phile and have a hatred for noise and recognize that robust analog design is difficult. Noise hurts my soul.

 

That was a fantastic post, thank you for your insights! Speaking of noise, this is a great example of what makes SBAF great, amazing signal to noise ratio.

Anyways, a clarifying question if I may; If your cable is short (less than a metre) and your usb source is clean, is there any real difference in noise received from the cable type?

Will a poor cable increase noise to any measurable degree over a short <1 metre run?

 

Everything is relative There will be a difference, given the same length cable, between higher quality and lower quality cables with Quality being defined by Wire Gauge, Wire material, Shielding, and Connector quality.

I tend to think of short cables in the < 1 ft range, though a standard 3 ft cable (~1 m) should be reasonable.

Even at USB 2 speeds, you'll see a signal attenuation difference between a really nice cable and a fair cable. I don't believe you should see much random noise difference though, unless the connectors or shielding are really poor. So the short of it is no, I don't believe you'll see enough difference with a short cable to justify the added expense.

The only USB implementation that I could readily look at is the Schiit USB boards they have on their website. Given how they have isolated the USB board, (and I assume power planes), and have used a quality USB chip, I'd be surprised that anyone would see much difference with cable or re-transmission (on the Bifrost and above) unless the USB transmitter in your PC is really spewing some crap. I would expect to see more of an issue with all-in one dac/amps with small footprints.

 

Thanks for pushing some facts into this debate! I do wonder if all this talk about re-transmission is a bit of a red herring. It let's DAC manufacturers off the hook for not investing in high quality USB implementations. I mean receipt of digital signal is one of the jobs of a DAC, one that should be integral to the design. There really should be no reason that anyone should ever have to buy a re-transmitter, what happens in the Regen or Wyrd should be done in the DAC itself!

I wonder what your thoughts are about the regen:
http://uptoneaudio.com/products/usb-regen

Is there anything it does that could be done equally well inside a DAC?

 

The consumer in me tends to agree. Though I do have some compassion for the designers.

When implementing USB for a keyboard or mouse, it's not going to matter much as long as the data is being received reliably. So foreseeing an interaction with the incoming USB data and the analog conversion wouldn't be something that is obvious to deal with in the design. As a designer, one would generally expect that the guy who designed the upstream component (in this case the USB receiver), created a working solution that doesn't interfere with you. Getting the USB data isn't the hard part of the DAC implementation, so one would expect to be able to grab something off the shelf to do it.

These kinds of noise issues would be found in a robust validation or characterization. Then if the issue could be isolated in debug (difficult), there would be a rev spin to account for it. Often, it wouldn't be possible to reproduce it or you would say, "it's not my problem, it's the Transmitter" or perhaps "most people won't be able to tell, so let's ship it". These are tough business decisions for small companies. Ship it, take the risk to make revenue and make payroll or delay for a rev and possibly go under while delaying the product)

Given that there aren't too many DAC's being made by large design teams, or with well equipped debug teams, it doesn't surprise me that such an complicated error mechanism would exist in some designs. It's a hard problem to foresee and a hard problem to isolate when you see it.

Every analog designer will have horror stories of some obscure bug that you would never believed was possible had you not seen it first hand.

As far as the regen is concerned, I should note that I haven't used it.
It looks to be correcting both poor transmitters and potentially some cable issues (though if you have the same crappy cable after the regen, you could still affect things). If it does what it claims, it will solve bad transmitter problems. Whether it actually successfully does what it claims I can't say.

It says it cleans up the signal and hands it to the DAC USB receiver in pristine condition, minus noise from the source Transmitter or power. That's a win in my book. The wyrd should do the same thing whether trying to or not.

The real inconvenience here is that its very tough for the end user to tell if their USB output is noisy or not. I know my USB ports on my PC are driven by the AMD SB850 southbridge and the NEC PD720200 chips on my motherboard. Quick googling reveals zero noise measurements attributed to those chips. So unless I scope it myself, I have to rely on whether I can hear the difference. I won't know if I can hear the difference until I hear it with and without a solution, which generally means I've spent the money already.

It's then a battle between your trust of your USB transmitter vs your trust of the DAC mfg to implement a robust design.

So I will probably buy one once my new gear comes in and see if I can hear any difference. I don't particularly trust the AMD SB850 design or the 3rd party NEC solution. I may have a different opinion about other chip manufacturer's parts. I do believe the Bifrost and later Schiit implementation looks promising, which lowers my concern for my own particular gear.

I know of 3 solutions that I'll likely consider: the Wyrd, the Regen, and the iPurifier 2. I like that the iPurifier 2 has no cable between the it and DAC. I like the look of the Wyrd (emotionally important). The regen is the most expensive of the 3, so I'll likely only go that route if the previous 2 fail to meet the need.

 

Absolutely! Any of the above solutions could be implemented inside the chassis. There is an added expense for that though., both in components and real estate. So the DAC companies are likely assuming that only a small percentage of people actually need the solution or will detect the problem. As such they are better off selling the cheaper DAC and then selling the stand alone solution separately for those that need it. There is a side benefit as well in that they can sell a product to "fix" issues on someone else's DAC, which could drive sales at a later date.


I should note that it would surprise me and disturb me greatly to discover that USB transmitter noise or USB power noise was a widespread problem. I haven't characterized any designs I haven't been involved with. I would expect any USB 3.0 implementations to be very clean. You can't achieve high data rates in a noisy environment, particularly across large distances. Though the USB 3 implementations that I am familiar with use different circuits when in USB 3 mode vs USB 1/2 mode.

 

I was looking at the manuals (because I find manuals more informative than sales blurb!) for Emotiva DACs yesterday. I was quite surprised to read, paraphrased, that some "audiophile" cables do not even meet USB standards and could detract from the performance, and the user is advised to buy a normal cable from a good manufacturer.

Well, the fact (if it is a fact: I'd love to see some of these cables measured) doesn't surprise me at all.

Even though I am a fan of the Archimago "C" USB cable, I'd think it crazy to wire one's stuff up like that except for purposes of experiment. Just a simple, to-specification cable is required --- but if people want to pay extra for pretty, then I'm more OK with that than I am with paying extra for psudo-science or the stuff that might, just possibly, just conceivably, go wrong.

On the other hand, I have thrown my any-cable-will-do philosophy out, due to discovering that simply, any cable would not do, just for charging my new phone at a decent speed: it requires the heavier-gauge power line. Not an audio, or even, directly, a data issue, but with many small DACs being USB-powered, could be a realisitc issue.

 

Is using USB cables beyond the length of 3ft or 1m a NO NO? Does the same apply to coaxial cables? What about RCA? I'll try to stay below 1,50m for RCA and 1m for USB but when I think about adding a WYRD to the Modi Multibit it will result in having 2x1m (3ft).

 

Some say you should use 3FT or less, I use about a 6ft cable. Speeds are the same for me. The article I read could be incorrect so these are not my words, however speeds on USB 2.0 cables should maintain speed at a maximum length of 16 feet and 5 inches. Better yet, let me just quote what the article wrote (like I said, I have no issues with a 6 foot cable, others might, and this is not my article so I cannot validate the accuracy of the statements.:


"The USB 1.1 Standard specifies that a standard cable can have a maximum length of 3 meters with devices operating at Low Speed (1.5 Mbit/s), and a maximum length of 5 meters with devices operating at Full Speed (12 Mbit/s).

USB 2.0 provides for a maximum cable length of 5 meters for devices running at Hi Speed (480 Mbit/s).

The USB 3.0 standard does not directly specify a maximum cable length, requiring only that all cables meet an electrical specification: for copper cabling with AWG 26 wires the maximum practical length is 3 meters (9.8 ft)."

https://en.wikipedia.org/wiki/USB

 

I guess I'm a little bit late to the party. Couple of videos that explain cable properties to some extent, atleast as a lump parameter (especially valid for high frequency pulses like usb signals).

 

Thank you for the comprehensive post. It looks simple in words, but only when you get around to the teams that work on characterization, circuit design, simulation and validation, we really get to know how complicated stuff is, and often not understood well enough, beyond a level of abstraction.

Also I assume the timing at which the buffer fills might also stress the i2s generator following it, am I right?

Are there any books you could recommend to read these transistor non linearities in detail?

Describing the sonic changes in these is hard since it doesn't manifest as a simple bass boost or treble boost or midrange wonk. It shows up as a complex modulation. For example, my supra usb cable makes my dac sound as if phase is inverted, the bass feels bootleggish and there's a characteristic shout I cannot describe as a simple midrange boost. All while having better detail and clarity than the stock cable otherwise. I almost believed there was some bitflip happening lol, but considering the frames come with a crc check I doubt it. A complete frame drop is possible and I've experienced it even with latency issues on software side, again not sure on the cable side since the sound never cut off audibly. My uptone uspcb cable makes my dac even more detailed but without any of those aberrations of supra, hence sounding both effortless and detailed. I understand some DACs might be impervious to these, either because what causes these changes are compensated, or they just mask these effects under other consistent aberrations.

Just a note, can we have a subjective thread for usb cables, provided people describe their pairings properly? Will help anyone with similar setup. I really hope it doesn't go into derail mode, or hyperbole mode though. (And definitely no random condoscending)

 

There’s a digital cables thread. As you say, don’t go crazy with your descriptions of the differences you’re hearing though, or ridicule will no doubt follow:

https://www.superbestaudiofriends.org/index.php?threads/digital-cables.6860/page-5

 

In case @PerfectAnalog isn't still following this thread five years after his last post...

The answer is probably no, but what do you mean by "stress"?

There's a lot on the web -- just search for "transistor linear region". Any explanation that delves even slightly below the surface of the subject will look dauntingly math-heavy, but it's not terribly complex math.

So I'm a huge believer in subjective evaluation of audio gear: If a USB cable sounds better to you, it IS better. But I wouldn't try too hard to pin down exactly what physical/electrical behavior is causing the perceived difference between two USB cables.

Absolute phase, modulation of the analog signal, frequency response, digital data integrity, etc., are easy to measure. From my experience -- I looked at a LOT of PHY eye diagrams when my team was developing the first USB 2.0 microcontrollers and hub chips twenty years ago -- I'd predict that if you did measure those characteristics, you'd see a) no difference between one USBIF-approved cable and another, and b) no improvement in any relevant behavior of a non-approved cable over an approved cable.

Which, again, is not to say that USB cables can't sound different. It's just that if they do, I don't think it'll be because of any measurable electrical difference.

 

I have been using a Belkin Gold cable since I first hooked a dac to my laptop. 3 meters. It works, no blips, burps or farts. I have walked on it, tripped over it repeatedly, yanking it out of the socket. Even ran over it a few times with my chair rollers. I'd buy another, just for a spare. The USB that came with my Pulse Infinity is hooked to my printer. It works reliably, which is more than I can say for the Infinity.

 

Thank you very much. By stress I meant either causing a significantly different power consumption/current draw profile, and/or additional noise/skew imparted to the next link in the chain.

I wanted to learn more about transient properties of transistors, including stuff related to shot noise. There was a book on 1/f noise if I am right, haven't went through it yet. The transistor linear region, as the name suggests, is a region of linear performance, under certain assumptions right. I wish to learn the behavior beyond those assumptions.

An analogy to this would be learning transmission line modelling as a normal lump parameter, vs learning power line harmonics additionally. The latter adds another dimension to the study and makes the model closer to reality.

 

@fastfwd I have another doubt, but this time regarding amplifier measurements. Are R load measurements the same as measuring through an actual headphone. A real headphone gives a bit of "resistance" (not the electrical term) to the force of an amplifier by virtue of it pushing a mechanical load (air/diaphragm), it's almost similar to a motor interacting with power line. A resistor just gives electrical resistance, no back emf or anything of that sort.

I recently came across this video. . Starting from 7.24. I tend to see an analogy to amplifiers here. The no load test for the drill /batteryfeels similar to r load in amps and an actual loaded test feels similar to an actual headphone test in that it is now trying to do something mechanically using the electrical energy.

Hopefully the answer is simple. I was trying to think this relates to line harmonics discussed above too. If not and it looks to derail this thread, I'd like to take it to another thread.

 

I see. No, the FIFO between USB and I2S will smooth out the bursty USB data. Unless the incoming data rate varies enough to completely fill or starve the FIFO, the I2S logic will "see" the same input regardless of the FIFO fullness, so its behavior will also be the same.

Right. I think you'll find more specific information about the device physics by searching for "transistor linear region" than by searching for "transistor nonlinearity". But sure, try them both.

Yes, and "closer to reality" is important. But a solid understanding of the basics is even more so. Just as real-world physics problems are almost always about friction, and yet every high-school physics lesson begins with "Assume a frictionless surface...", the higher-order semiconductor phenomena will be easiest to understand if one already has an intuitive understanding of the first-order behaviors.

 

That is correct; a constant resistive load is not a perfect simulation of the varying impedance that an amplifier will see when it is driving a real headphone.

Vaguely similar, if I am understanding you correctly. But remember that the R load IS doing something physical: It heats up when a current is passed through it.

 

Thank you. You've understood me mostly correctly, but I wasn't much into varying impedance correlation (which can be deduced by relative gain equations on a conventional topology, using output impedance), as much as I'm into the concept of back emf. Joule heating is common to both resistors and also to headphones, pretty much everywhere current passes. Back emf and mechanical energy conversion is exclusive to headphones (the basic f = BIL equations and so forth), and that was what I was equating to the loaded test of the drill, the current consumed shoots up and can stress an inadequate battery. One more thing I'm thinking along the lines of, is inrush/transient currents. Not sure if it's of any significance in headphone transducers.

I'll do my search and see if I can find better inferences relating to cables and transmission line networks.