Great USB audio summary. 

"Most modern DACs with USB input implement "Isochronous transfer mode with asynchronous feedback".  The isochronous transfer mode of USB is the one that guarantees transmission bandwidth but does not support data retransmission (in response to data packet CRC errors).  In contrast, the bulk transfer mode of USB (the one used for USB flash drives & other storage devices) has a packet retransmission capability but no guaranteed throughtput.  Isochronous transfer mode favors data transmission regularity over integrity, whereas bulk mode is the opposite.

In USB, strictly speaking there is no asynchronous mode, but rather "asynchronous feedback", so for the DACs, it is not a choice between implementing isochronous or asynchronous.

Asynchronous simply means the receiver (being the USB DAC) will send information back to the source (PC, streamer, etc.) to slightly speed up or slow down the data transmission, to prevent buffer underruns or overruns at the DAC side.  Essentially the USB DAC controls the pacing of the data.

With good USB signal integrity, achieved with good quality and not-too-long USB cable, data corruption should be a non-issue.  The isochronous transfer mode with asynchronous feedback is the correct way to send audio over USB, and tremendously helps prevent audio dropouts due to interruptions in the data stream."

Scan80269 as quoted from audiostyle.com, April 2017

@thyname I have 50 years experience in audio, professional, home theater and 2-channel., and I am formally trained in architectural acoustics. several of my professional speaker designs have been commercial successes. I also have 40 years experience computers and digital audio, including 20 years as a principal network architect for AT&T.

I have never said analog cables do not make a difference, but claims made for digital cables simply have no plausible or viable basis for their claims. The nature of digital signal transmission and the protocols used do not support the claims. Whatever audible differences that may exist are in the ADC/DAC process involving quantization errors or sampling rates, or in filter algorithms or improperly filtered noise on the analog outputs. None are attributable to the digital data transmission itself. A quick study of the layered model of data communications, like the 4-Layer TCP/IP model shows why this is the case

The TCP/IP reference model has four layers:

4) Application Layer - Formats messages, provides User Interface, and App Services

3) Transport Layer or Host to Host Layer - Ensures data delivery and sequencing

2) Internet Layer - Provides addressing and routing

1) Network Access Layer or Link Layer - Provides physical connectivity and transport of raw bits.

Layer 1 - The Link Layer is where the physical connection lives, be it copper or glass or radio. Layer one serves only to transport raw data bits - low voltages (0s) and high voltages (1s) or in the case of fiber, relative light and dark levels. This where all digital cables operate and is the only physical component. Everything above this is software. The actual data rides above Layer One, in the case of Ethernet and USB, in packets.

Transmitted signal levels are 0.0–0.3 V for logical low, and 2.8–3.6 V for logical high level. This means that any 'noise' below 300mV is simply not recognized. Any noise above that, and that would be a VERY noisy circuit, would trigger a CRC or Parity error, and a packet retransmission for TCP/IP.  It's also easy to measure with an oscilloscope. 

You should also be aware that USB 3.0 (the standard since 2008) and above are operating at 5 Gbits/Sec, 10Gbits/Sec for 3.1, 50,000 to 100,000 times faster than even the most aggressive audio requirements. Discussions of isochronicity in relation to audio signals are academic at best for any realistic implementation. So if you want to run a 100M or 200M USB cable, yeah, maybe expect issues. At 5M or less, just not a big deal, not even a little deal.

You can listen all you want and perceive all the differences you want, but they aren’t in the digital cable.

@soix If a cable is useful in mitigating noise, then the issue in the noisy component, not the passive cable. If a cable with a telescoping ground (grounded at the input, but not the output is necessary to minimize hum, then the problem is a ground loop elsewhere, and the cable only a mitigation. In either case the perception of the cable improving matters is incorrect. It is only covering up the real, and typically worse problem elsewhere, whether it's 3 semis of sound gear for a show that night, a 20 megawatt data center that's blowing PDUs, or a hifi, the fundamental issues are unchanged. If a cable makes an audible difference, or changes increase packet drops, the root problem lies elsewhere.

@atthyname Actually, it does qualify me as an expert,, seeing as how AT&T invented the opamp, the venerated 300B tube, the transistor, and yes, digital audio: Digital Pulse-Code Modulation was invented at Bell Labs in the 1930s and first used as a telephony technology. In World War 11, the military phone line between London and the Pentagon was compromised and the Germans were able to break the non-digital security system. Engineers at Bell Labs developed a PCM-based encrypted-transmission system called SIGSALY, which was deployed in 1943.' The system eventually grew to 12 terminals before being retired in 1946. Patents on the 12-channel encryption system were classified until 1976. SIGSALY represented the first digital quantization of speech and the first PCM transmission of speech. 

But, jumping forward a few decades, who exactly carries the data from your digital music provider to your local ISP? Most of the time it's AT&T. On a massively scaled global IP network, carrying ... Zeroes and Ones. They don't discriminate between voice, text messages, a spreadsheet, e-commerce, streaming video, and our topic here, music. By the very nature of their architecture, at Layer One they do not care, they are not even aware of the nature of their content. It's all, it is only Zeroes and Ones. Coming from an analog background, it took me awhile to realize the implications of such a democratic technology. But is is an inarguable truth. The physical layer has no construct that would allow it to discriminate between data types. Or have any impact on the data quality whatsoever. All those decisions are in the multiple software layers above. One last example. Does it matter if your groceries are delivered to you local supermarket in a Kenworth or a Freightliner? No, of course not. The quality of your food is determined by a myriad of other relevant  factors. If it arrives spoiled, do you blame Peterbilt? If a truck carries the best steak you ever had do you credit Volvo? They picked it up at the source, transported it to the destination as fast as the law and the traffic allows, and it was offloaded for additional processing, delivery, and your enjoyment. 

So it might do you both well to get your facts straight, avoid the ad hominem attacks and actually learn a thing or two about digital communications.

@jerryg123 Eexpertise in rocket science? I don't know, you'll have to tell me. 

I do know I brought my technical experience in analog to my work in digital communications and networking, where it served me well in developing high bandwidth and high capacity network architectures, especially in paying attention to grounding and noise management in high density rack installations, beginning with T-1 (1.544 Mbps) and T-3 (45 Mbps) in the early 90s to OC-12 (622 Mbps) circuits in the early 2000s. And by 2008 AT&T had upgraded over 80,000 miles of carrier backbone to OC-768 IP/MPLS 39 Gbps. About the I changed roles developing high-cap data replication, but now most of the network is100GbE moving towards 200 and 400.

And that's my point. No matter which music streaming service you're using, no matter where in the cloud (or rather which hosting services are used), there's a shit-ton of networking and thousands of miles cabling, both copper and fiber involved, none of it anything more than regular commercial grade. Swapping out the last 72" will give you better zeroes and ones? 

@thyname When you are dealing with Layer 1, the Physical layer where cables dwell, there is no such thing as 'digital audio transmissions'. There are only 0s and 1s represented by 0-300mv for Zero and 3.6-3.8v for Ones. Layer 1 has no mechanism for discriminating any data types, that all occurs at Layer 3, 4, or above, and is software, not hardware. 

So how does a cable give you better 0s and 1s?

@thyname "Yes.  Just like interconnects, speaker, and power cables can make a big difference.  Rather than being a digital cable denier I’d suggest you try a better digital cable and use your ears rather than theory to make any conclusions.  BTW, what digital cable are you using now?"

Your statement conflates analog and digital communications So how are the zeros and ones made better?

No one on this thread has yet to give any plausible explanation for how a Layer 1 cable impacts sound quality. 

There are lots of things that can impact SQ in the analog realm that possibly relate to cables, but none of those apply to digital. It is in the very architecture of digital communications that this is so. Analog and digital communications simply function entirely different ways. This is not a matter of discussion; it is physical fact.

Just like there is no such thing as quantization error in analog; there are no such things as RF rejection, or ground loops in glass fiber. And at Layer One in digital communications, whether copper, glass fiber or RF (WiFi and Bluetooth) there are only zeros and ones, regardless of whether it's audio, video, social media or e-commerce. That's all there is, nothing more. Until you grasp that simple concept, you simply cannot claim any understanding of digital communications, or what may ultimately impact sound quality exiting a DAC.

@thyname "And please enlighten me, where else in cables we should look for a difference maker?"

You shouldn't be looking at cables at all, you should be looking at Codecs, more specifically how a given codec has been implemented, and the containers used.

And the containers

With even a quick browse you'll realize there's lot of ways these algorithms can be and are implemented, and coding one way and decoding in another can lead to all kinds of spurious output issues.  

Oh, and as for my personal listening, I stream Amazon UHD because Prime and Qobuz because it's probably overall the best.

My DAC uses the AK4490 DAC chip set to an unstressed 24-bit / 192 kHz (or DSD 5.6 MHz) in deference to the extra noise likely to be an issue at the full 32-bit / 768 kHz (DSD 11.2 MHz) implementation.

My preamps include a modified and upgraded APT-1 and a Marantz PM7000N.My power amps include the Nelson Pass designed Adcom 545 100W/ch MOSFET and a Mcintosh MC240.

For speakers, I have KEF LS-50s, Magnepan LRS, and Monitor Audio Silver 300 7G, as well as Grado and HiFiman Sundara headphones.

Other Sources include a VPI Prime Scout w/JMW-10-3D Unipivot arm and Hana SH moving coil and a Marantz CD6007 CD Player. 

While we’re at it, here’s another good reference

Digital to Analog Conversion

https://www.sciencedirect.com/topics/engineering/digital-to-analog-conversion

"After DAC, the analog signal is sent to the anti-image analog filter, which is a lowpass filter to smooth the voltage steps from the DAC unit.". There are numerous filter designs and theories about them. I leave it to you to conclude the best.

Concerning quantizing noise, there are four kinds: overload noise, random noise, granular noise, and hunting noise.

Overload noise is the level of analog waveform at the input of the PCM encoders must be set such that its peak value does not exceed the design peak of the quantized V volts. If the peak input exceeds V, then the recovered analog waveform at the output of the PCM will have some flat tops near the maximum values.

Random noise, This noise is the result of random quantization errors in the PCM system under normal operating conditions.

Granular noise When the analog input level is reduced to a relatively small value with respect to the design level, the error values are not equally likely from sample to sample.

Hunting noise This occurs at the output of a PCM system. It can occur when the input analog waveform is almost constant, even when there is no signal during these conditions.