Empirical Audio Off-Ramp 4 USB format converter Measurements

Sidebar 3: Measurements

I measured the behavior of the Empirical Audio Off-Ramp 4 using the Audio Precision SYS2722 system (see www.ap.com and "As We See It" in the January 2008 issue). I installed the correct version of the driver program for the Snow Leopard operating system, and played test tones at various sample rates and bit depths from BIAS Peak 7 on an Intel MacBook running OS X 10.6.8. The Macintosh USB Prober utility reported both the manufacturer and product strings as "Empirical Audio Async 192." The interface was identified as "vendor-specific," with no information offered about the number of operating mode, channels, bit depth, or compatible sample rates, presumably because these are all aspects of the driver program.

The Off-Ramp 4 operated at sample rates of 44.1, 48, 88.2, 96, 176.4, and 192kHz, provided the sample rate had been set with Audio MIDI Setup on the Mac and its output correctly followed the bit depth of the file being played, 16 or 24. Most important, the output bits were the same as those in the file.

Fig.1 shows the eye pattern of the AES/EBU output, the Off-Ramp 4 driving the digital input of the SYS2722 with the 1.5m length of Kimber Orchid cable I used for my auditioning. The eye is wide open, with no blurring at the ends that would indicate the presence of significant amounts of jitter. The Audio Precision calculated the level of jitter to be just 343 picoseconds peak, over a 50Hz–100kHz bandwidth. For reference, the BNC version of the Halide Bridge's S/PDIF output produced 440ps peak measured over the same bandwidth (footnote 1); the Musical Fidelity V-Link's output measured 395ps peak. This measurement was with the AudioQuest Coffee USB cable and the Off-Ramp powered from the Monolith's battery; the jitter number didn't change when I switched to the wall-wart supply or to a generic USB cable.

Fig.1 Empirical Audio Off-Ramp 4, eye pattern of AES/EBU data output carrying 16-bit J-Test signal (±2V vertical scale, 175ns horizontal scale).

Fig.2 shows spectral analyses up to 20kHz of the timing uncertainty in the Off-Ramp 4's AES/EBU output (red trace) and the BNC Halide Bridge's S/PDIF output (blue), both performed in the digital domain while the datalinks were carrying 16-bit J-Test data. While the absolute levels of the spectral components are all low, at 20ps or below, the spectra are very different. Though it has a relatively strong component just below 19kHz, the Halide's spectrum is very clean. If you look closely at the blue trace, you can see very low-level components at 10,796 and 11,254Hz, these equivalent to the frequency of the Fs/4 J-Test tone ± the frequency of the Fs/192 LSB-level squarewave. These are both accentuated with the Off-Ramp 4, with higher-order components visible (red trace). When the AES/EBU datastream is fed to a D/A converter, these higher-frequency timing variations will be reduced in level by the low-pass filter action of the DAC's Phased-Locked Loop receiver circuit. I have shown them here to indicate that the Off-Ramp's jitter signature is very different from the Halide's as well as from the Musical Fidelity V-Link's, which was noisier across the spectrum (not shown). Interestingly, the Off-Ramp's spectrum was slightly cleaner at higher frequencies with the generic USB cable than with the AudioQuest Coffee.

Fig.2 Empirical Audio Off-Ramp 4 with Kimber Orchid 1.5m AES/EBU cable (red) and Halide Bridge, digital-domain jitter spectrum of datastream carrying 16-bit J-Test data (linear horizontal frequency scale, DC–20kHz, logarithmic vertical scale, 1ps–2ns).

As in previous reviews of USB converters, I examined how the Off-Ramp 4 affected the analog output of the D/A converters with which it was used (footnote 2), in this case the Logitech Transporter and dCS Debussy. Starting with the Transporter, fig.3 shows the spectrum of its output fed a 24-bit version of the J-Test signal from the Off-Ramp 4 via the 1.5m length of Kimber Orchid AES/EBU cable. The floor is relatively noisy, and while the central spike that represents the 11.025kHz tone is sharply defined, sidebands are visible at ±229Hz and its harmonics. These shouldn't be there; obviously, there is something in the Transporter's receiver circuitry that is sensitive to datastream jitter. This graph was taken with the Off-Ramp powered from its battery supply; changing to AC power didn't affect the spectrum, and neither did changing from AudioQuest Coffee to a generic USB cable, or the AES/EBU connection to S/PDIF.

Fig.3 Logitech Transporter, high-resolution jitter spectrum of analog output signal, 11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 24-bit data from MacBook via Empirical Audio Off-Ramp 4 and 1.5m Kimber Orchid AES/EBU cable. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz (left channel blue, right red).

For comparison, fig.4 shows the spectrum of the Transporter's analog output while it was fed the 24-bit J-Test data via the Halide Bridge and the BNC connection. The ±229Hz sidebands are 6dB higher in level and the noise floor is now dominated by closely spaced sideband clusters. So while the Halide Bridge's datastream output has fewer timing variations, the Transporter appears to have more of a problem rejecting those timing variations than it does with the Off-Ramp 4.

Fig.4 Logitech Transporter, high-resolution jitter spectrum of analog output signal, 11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 24-bit data from MacBook via Halide S/PDIF Bridge with BNC jack. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz (left channel blue, right red).

Does this explain why I found the Empirical Audio converter to sound better than the Halide with the Transporter? Perhaps. I did find the differences between the converters to be much harder to hear with the dCS Debussy. Fig.5 shows the Debussy's output fed USB data representing 16- and 24-bit J-Test signals via the Halide Bridge. All that can be seen with 16-bit data (cyan and magenta traces) are the odd-order harmonics of the low-level squarewave, which lie at the residual level. Changing to 24-bit data eliminated these completely, though a couple of low-level discrete tones remain between 9.5 and 10.5kHz. Repeating the spectral analysis with the Debussy fed by the Off-Ramp 4 gave an almost identical result (fig.6), though the spectral spreading at the base of the central spike that represents the Fs/4 tone is reduced in both amplitude and width.

Fig.5 dCS Debussy, high-resolution jitter spectrum of analog output signal, 11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit (left channel cyan, right magenta) and 24-bit (left blue, rightred ) S/PDIF data from MacBook via Halide Bridge. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.

Fig.6 dCS Debussy, high-resolution jitter spectrum of analog output signal, 11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit (left channel cyan, right magenta) and 24-bit (left blue, right red) AES/EBU data from MacBook via Empirical Audio Off-Ramp 4 and 1.5m Kimber Orchid cable. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.

These measurements indicate that the Empirical Audio Off-Ramp 4 does a superb job of transforming USB audio data into an AES/EBU or S/PDIF stream. However, whether or not it offers an audible improvement with a specific D/A processor will depend on the processor's inherent rejection of datastream jitter.—John Atkinson

Footnote 1: When I reviewed the Halide Bridge, I measured a lower 343ps peak; I don't know why my new measurements gave a higher figure.

Footnote 2: A primer in why and how I measure jitter can be found here.

Empirical Audio
13852 Bishops Cap
Black Butte Ranch, OR 97759
(541) 595-1001

VandyMan's picture



Very interesting review. I look forward to seeing a comparison to the Alpha USB. I thought it was too expensive for a USB converter ($1600), but it looks like a potential bargin compared to the Offramp 4. 

thestewman's picture


Did I miss something in your article ?

"Up to this point I'd been using a generic USB cable with the Off-Ramp. I now substituted the expensive AudioQuest Coffee cable ($295/1.5m). If I heard a difference, it was very small, but if I had to swear on J. Gordon Holt's unwritten autobiography, I'd say that the AQ continued in the same direction the improvement I'd heard with the battery supply."

 "I spent the rest of that day trying to measure the differences between these two USB cables. It proved as frustrating as a snipe hunt, though I did find a very slight difference in jitter between the AudioQuest and generic cables with the Transporter: The generic USB cable was better."


The Transporter does not have a USB input. 

John Atkinson's picture

The comparisons were between the premium and generic USB cables feeding the Off-Ramp 4, which in turn fed the Transporter via AES/EBU.

John Atkinson

Editor, Stereophile

Johnny2Bad's picture

RE: " ... my 2006 G4 Mac mini runs OS X 10.5 (Leopard), the final version compatible with its obsolete G4 processor— ..."

Unlike the majority of current day Mac users, I am one of those people who bought a 68000 Mac running System 6.0.8 and have used Macs daily since that day in 1990. This means three incompatible CPU families (24 and 32-bit Motorola 680x0 series; 32 and 64-bit Motorola and IBM PowerPC series, and 64-bit Intel x86 series).

During that time I also worked extensively with WindowsOS and Linux / UNIX systems, including authoring documentation for a prominent Linux distro ( note that the typical Linux contributor would rather eat glass than write documentation).

Furthermore, I've been using the audio authoring and playback performance capabilities of these machines to the fullest as the current technologies allowed, and using the computer as a primary audio playback device in a High Fidelity Audio System for the last 20 years.

I can assure you that the Motorola/IBM G4 and IBM G5 CPU's and the associated MacOS Operating System code, third party software and Apple/3rd party hardware is not only adequate for the task, it in many aspects is superior to the Intel x86 hardware that supersedes it. It is only in the last 2 or 3 years that the x86 Macs can claim to catch up in Audio processing power to what G4 Macs could do in 2001.

I never experienced dropouts, not even once, while using G4-based Macs in real-time authoring or playback of up to 8 simultaneous 24/96 channels; it was a huge shock to discover dropouts on recorded audio files, after the musicians had gone home, of course, when reviewing with my first 2 GHz dual core Intel CPU-based Mac that, on paper, "ate for breakfast" the lowly 867Mhz G4 it replaced.

Naturally that led to further investigation on my part, and I can assure you the G4's are extremely competent audio performers (and video; you can play back12 Quicktime videos simultaneously, and actively manipulate any one video ... including scrubbing ... without a dropout on a G4 ... you might be able to simultaneously play back maybe 3 videos, don't attempt scrubbing, on the Intel x86 that first replaced it. There is much that was lost when moving from a PPC's RISC architecture to the x86's CISC when it comes to real-time data handling.

Obsolescent ... maybe. Obsolete ... hardly.