Gordon Rankin on why USB audio quality varies

The in-betweeners. The hifi gear review process is a little like USB audio transmission. Between product arrival and publication sits a process. What takes place during that process – we’ll call them the in-between days – influences the quality of the outcome. Mistakes made and short-cuts taken show up in the end result.

Not all of it is super-serious work. Playtime matters just as much. Skirting the periphery of review formalities, especially on weekends, I often prefer to play with the gear at hand. I try out alternative configurations and connectivity options. This experimentation is invariably the result of asking the simplest of questions – “What if?”.

The outcome of these loosely defined play-experiments often begins with a single line of enquiry. The result however often lands in an entirely different place. Occasionally, I defer to manufacturers or engineers for advice on the same. After all, if I don’t know the answer, chances are I know someone who might.

This is one such example.

In the wake of May 2016’s Munich High-End and Fujiya Avic Spring events, I began to ponder network streamers and how Roon Ready endpointing is slowly moving from external devices (like the AURALiC Aries Mini) to Ethernet-fuelled boards found inside DACs: PS Audio’s DirectStream Junior, Ayre Acoustic’s QX-5 Twenty and AURALiC’s ALTAIR are three examples that I caught in Munich.

Prior to leaving for the aforementioned month of audio show reportage I’d purchased a Sonore Sonicorbiter SE – a US$299 dongle-like streamer based on the CuBox hardware platform. Specifications included a multi-core ARM processor, 1Gb RAM, Ethernet input and USB/Toslink outputs.

On the software side, the Sonicorbiter SE runs a customised version of Linux that allows for switching between a handful of operational modes, chief among them Squeezebox emulation, Roon Ready endpointing and HQPlayer NAA.

With the Sonore cube offering DSD output up to 512 times the rate of CD’s 44.1kHz, DSD512 compliance is provided natively or via DoP. Which one works for you will depend on the DSD implementation inside your DAC. Connected to my Sonicorbiter SE via USB was Mytek Digital’s new Brooklyn (US$1999). It goes native.

Here our story finds a fork in the road. HQPlayer opens up the possibility of on-the-fly PCM to DSD up-sampling that takes place on the server. The resulting DSD is then streamed across the network to the Sonicorbiter running in HQPlayer mode. Thoughts on this will feature in another post.

Then came a second question: what if the Sonicorbiter SE could be configured so that it approximated the sound quality of the AURALiC Aries Mini running as digital streamer (with stock PSU), also connected to the Mytek DAC via USB?

The Sonicorbiter’s trump card is Roon Ready – something that AURALiC CEO Xuanqian Wang has obliterated from the development roadmap of his entry-level streamer. “Roon Ready is never, ever, ever, ever EVER coming to the Mini!” he told me in Munich. You got that?

In stock form, even with Curious Cables’ most excellent USB lasso, the Sonore cube sounds a little lacklustre with micro dynamic flair and tonal colours come across as a little bleached. Experience tells me that this is the result of electrical noise polluting the DAC and/or jitter; or the former causing the latter.

I considered applying corrective measures at the source. The Sonore is powered by a fairly ordinary switch-mode brick. Sonore’s linear option is 110V only whilst iFi Audio’s iPower is touted as a superior switch mode option for those living outside of the USA. I didn’t have have access to either PSU.

What I did have to hand was iFi Audio’s USB iPurifier 2 (US$109). What if bits weren’t ‘just bits’?

The iPurifier 2’s designer and iFi mainman Thorsten Loesch pitches his USB dongle as follows:

“One of the main bugbears of computer audio is EMI noise. The noise is generated by the raft of generic switch-mode power supply units scattered throughout the computer being in close proximity with the audio path. This is an unwanted distraction to the point of being downright annoying. The residual noise floor is heard most patently between tracks.”

The implication is that even with a quieter power pack fuelling the host device (e.g. Sonicorbiter SE), internal switching devices inside the host can cause EMI noise to spill from its USB port.

iFi Audio’s website text again: “Taken directly at the computer’s USB port, EMI noise is typically some 39dB. With the iPurifier in place, this noise drops by some 5dB to 34dB.”

Having already tested and written about the Schiit Wyrd, the AudioQuest JitterBug, the UpTone Regen and the Wyred4Sound Recovery, extensive real world experience tells me that USB signal cleaning devices like these can make a noticeable difference to sound quality.

Like the v1 edition (reviewed here), the iFi iPurifier 2 connects directly to the DAC’s USB port and cleans up the USB signal after it has left the host device’s USB port and journeyed down the USB cable. In this sense it is most similar to the UpTone Regen. However, the iFi requires no intervening cable or dongle to make the DAC connection; it plugs directly into the port.

The iPurifier 2 is also an active (powered) device yet there is no external power brick feeding it. Instead, it gets it go juice from the host device (computer/streamer) – in our case the Sonicorbiter SE.

The iFi dongle brings greater tonal and micro-dynamic avidity to the Sonore table. The before and after somewhat reminiscent of the day I first added a S/PDIF re-clocker to a Squeezebox Touch.

The upshot – and the answer to this hardware experiment’s “What if?” genesis – is that the Sonicorbiter + iPurifier 2 pulls up alongside the AURALiC Aries Mini’s performance as a digital streamer. Nice.

Here be monsters! For those thinking they’ll be quids with the Sonore/iFi combo are reminded of the need for a server (running Roon Core or Squeezebox Server) elsewhere on the LAN, even for cloud streaming. The Aries Mini and its associated Lightning DS control app can run solo with Tidal and can accommodate an internal hard drive for local library content. The Aries Mini also comes fitted with its own very good internal DAC.

The Sonicorbiter SE strikes a different chord in the market – it’s a digital streamer for those already in possession of a Roon/Squeezebox server and DAC.

Of greater interest to me was why the iFi dongle would make such a marked difference? Many readers baulk at the idea of deluxe USB cables like the Curious and USB ‘signal integrity’ cleaners like the iFi iPurifier 2 making the slightest (errm) bit of difference. Bits are bits are bits is their seemingly well-reasoned argument.

Keep in mind the financial safety (no upgrade required!) maintained by consumers who adhere to the simple notion that digital audio transmission over USB is all ones and zeroes, that bits are just bits and as long as those bits all arrive in tact, everything should the sound the same because buffers.

But no.

To learn why (and to see proof), I contacted Gordon Rankin via email. Rankin is the founder / chief engineer at Wavelength Audio. He introduced the digital audio world to asynchronous USB transfer – now the accepted standard – and he has worked on the USB implementation found inside Ayre Acoustic’s QB-9 DAC and, more recently, Feniks Audio’s active Essence loudspeakers (among many, many others).

Rankin is perhaps best known for his work on AudioQuest’s DragonFlys; this year’s Red and Black are the most recent implementations of the Ohio resident’s USB know how. In other words, Gordon Rankin is, to my mind at least, Mister USB Audio.

Who better to ask then: why don’t all USB audio configurations sound alike?

Via email, I posed the following question to Rankin: when I transfer a file over USB to an external hard drive it doesn’t make transfer errors – the file at the destination is the same as the source – so why should sending digital audio over USB be any different?

What came back was an epic reply. Strap yourself in – we’re going for a ride.

“While all of these interfaces (Firewire, SPDIF, USB, Ethernet, Thunderbolt so forth) have specifications. The % differential from one supplier to another in electrical, cabling, device and host seem to vary quite a bit.“

“All data moving between a host computer and a device over USB is done electrically. There are different speeds and different protocols that determine how a device and the host communicate.”

“Any interface between two points cannot be totally error-free. If you use a hard drive over USB, Ethernet or Firewire there are transmission errors. That means the transmitting device is told to resend the packet that has the error in it. Most of the time this is one bit in a packet size of length X.”

“Remember, the carrier is modulated on the data so the larger X, the bigger chance of errors. Also the faster the interface the more chance that there will be an error.”

“The three main USB transmission protocols are Bulk, Interrupt and Isochronous. Bulk (used for data transfer to a hard drive) and Interrupt are error-correcting. Isochronous (used for audio) is not.”

“Bulk and Interrupt are immediately NAK (negative acknowledgement). The receiver is designed to detect a bad packet immediately and the packet is resent.”

“For USB audio, the receiving device is basically translating a serial stream of data with a clock interwoven throughout. At the end of the packet sits some sort of block check. If the block check does not match the data then that packet is flagged as an error.”

“With Isoschronous USB transmission, packets are sent without any error correction / resending. But guess what? This is the USB protocol used for audio frames. The bad news is they are not error-free. The good news is these Isochronous frames are afforded the highest priority in the system.”

“A couple of years ago, I bought an expensive Tektronix USB setup. I have had protocol analyzers since designing my first USB DACS some twelve years ago. The Tektronix is useful because it allows me to see errors better both in electrical and data packets.”

“The big thing that many people don’t realize is that not all USB ports are created equal. Not all USB cables are created equal and it’s the same for devices and even operating systems. Since getting the Tektronix I have tested probably thirty different USB cables on the fifteen computers in my lab. These computers run a variety of operating systems and the Tektronix results vary between computers even when the cable remains the same. Let’s just say it’s not as pretty as I thought it would be.”

“Just a couple of things to think about in regards to USB ports. First, look to see what else is located on that tree. Each USB port can handle 127 devices. Sometimes there are additional ports hidden (inside your computer) and there are internal devices sitting on those ports – this could be the same tree that is hosting your USB DAC”.

“You can see in the lower tree, USB Hi-Speed Bus, Hub*, Hub, Apple Internal Keyboard/Trackpad, BRCM20702 Hub, Bluetooth USB Controller. On some computers, Hub* will actually be an external port and if you plugged your DAC in there you’re sharing it with all the devices below it.”

“On the PC, you can use this helpful application from Thesycon.”

“Alternatively, if you know your way around Device Manager you can go through that to find the USB tree. Although neither of these really give you a good indication of which port is on which internal hub, if you ARE able to place your DAC on a direct port you will be best served.”

“Speed plays an important part in all of this too. You may have heard the terms UAC1 and UAC2 – these are USB Audio Class protocols. UAC1 was designed for Full Speed device and host interaction. A data packet is sent every 1ms. In that packet are up to 1023 frames.”

“In high speed or UAC2 those 1024 frames each contain eight micro frames. Therefore, the amount of data we can send over UAC2 is basically eight times greater than that of UAC1. But with more data at faster speeds comes more errors and system configuration becomes harder. I almost never see an error on a UAC1 device, on a UAC2 device I can pretty much count on errors in both directions.”

“For optimum results, at least in theory, it’s best not to use a USB hard drive for your library with a USB DAC connected to the same host device. Think of it this way: your music software is reading from the hard drive in a synchronous manner and then writing to the DAC in that same synchronous manner and, as the DAC has priority, the music software might fault when reading the disk – this can lead to really bad sound.”

“Also, it’s probably best not to put the library on the system disk – because system stuff has really high priority over music playback software and again the music software can fault and bad sound will result. When a music app faults it becomes NON-bit true. One workaround for this is to choose a music app with memory buffering but in my experience, even that’s not guaranteed to be 100%.”

“A good example of this is when we transitioned from Full Speed USB to High-Speed USB DACs. A lot of the really expensive USB cables from audio companies failed miserably; I doubt many of these cables were even tested for High-Speed compliance.”

“To summarise: the problem with USB Audio is that Isochronous USB frames are not error-correcting. Therefore the sonic outcome of any USB system is dependent on the host to device differential.”

“Twelve years ago, I pretty much thought as many people do today: that USB was the answer to our S/PDIF quandaries. In some ways, it is a good deal better. We have Asynchronous Isochronous so the device and host know about sample rates, bit rates, clocking options and a host of other things. But cables make a difference, computer brand and quality make a difference and even the device makes a difference.”

“I will try and capture some data errors on the Tektronix. The problem is that it does not accumulate errors. It also does not stop on errors. I have to actually push a button to capture it and then pipe that to my screen.”

“What this is showing is the event table or the decoding of the USB Bus as seen by the Tektronix scope. We are using a Tektronix Differential Probe and their USB Analysis package. I made a little Male/Female HS USB board and that is plugged into a MacBook Air. I am using the Faber Acoustics Signal Scope Pro to send a 1KHz Sine wave to a Wavelength Crimson DAC @ 176.4K sampling rate. This is a program I use to test basic DACs of all kinds. This is also a pretty basic setup compared to audio transmission.”

“As per the above, a packet sent by the host to the Crimson has a CRC16 error. You can see that in the error column.”

TL;DR? What we have here is an explanation with screenshot proof that USB audio transmission isn’t bit true – enough evidence to reject any null hypothesis that assumes 1) all bits sent will all arrive intact and 2) re-transmission sorts out any detected errors. The Isochronous protocol used for audio data checks for errors but does not do any error correction (by way of retransmission).

Bringing it all back home, the iFi iPurifier 2 likely improves the sound of the Sonicorbiter SE because it minimises transmission errors by making lighter work for the Mytek Brooklyn’s USB receiver chip.

Furthermore, with lower noise on the line, the USB receiver chip no longer needs to kick into a higher, noise-inducing operational gear to ensure that it reads the incoming data correctly.

In other words, USB audio isn’t simply a matter of bits leaving the host PC/streamer and arriving AOK at the DAC. The quality of what sits in-between matters.

Further information: iFi Audio | Sonore | Wavelength Audio