The γ3 high resolution DAC

Specifications

Input modes
Resolution (bits) 16 or 24
(auto detect)
Supported PCM sample rates (KHz) USB UAC1 mode: 44.1, 48
USB UAC2 mode: 44.1, 48, 88.2, 96, 176.4, 192
AES/EBU, coax, optical: 32-192
(auto-detect)
PCM up-sample frequency (KHz) 96 or 192
(user-selectable)
Digital filter
WM8741 internal 1. Linear Phase "soft knee" filter
2. Minimum Phase "soft knee" filter
3. Linear Phase "brick wall" filter
4. Minimum Phase apodizing filter
5. Linear Phase apodizing filter
(user-selectable)
Analog filter
All modes Low-pass filter, Fc = 100KHz, 18dB per octave, Bessel type
Digital inputs
Digital input 0 USB with asynchronous transfer mode (Mini-B socket, 5-pin)
UAC1 - full speed
UAC2 - high speed
(user-selectable)
Digital input 1-4 AES/EBU (XLR), S/PDIF coax (RCA or BNC), optical (Toslink)
(build-time configurable on a per input basis)
Digital outputs
Digital output 1 Optical (Toslink)
Digital output 2 AES/EBU (XLR), S/PDIF coax (RCA or BNC), optical (Toslink)
(build-time configurable)
Analog outputs
Types balanced (XLR)
unbalanced (RCA)
Output voltage
0dBFS, maximum volume
4V RMS (balanced)
2V RMS (unbalanced)
Number of channels 2
Load impedance Can drive loads down to 300Ω
without significant degradation of performance
Power
Source 120V or 240V AC mains, 50Hz or 60Hz
Software
USB host operating system Linux, Mac OS X, Windows, iOS, Android

Note: γ3 will not decode non-PCM audio or multi-channel data streams.

Measured results

All measurements were performed on AMB's reference γ3 DAC. The tests were done using the following equipment:
  1. Audio Precision System Two SYS2322 Dual-Domain audio analyzer
  2. Altor Audio JK-GEN 384 digital signal generator
  3. Tektronix TDS2014B 100MHz quad trace digital storage oscilloscope
  4. Fluke 45 digital multimeter
  5. Fluke 187 digital multimeter
  6. Battery-powered laptop host computer, with signal generator and spectrum analysis software
  7. other custom test jigs, cables and software
Except otherwise noted, most tests were performed at 24 bit, 44.1KHz sample rate, via coaxial (BNC) S/PDIF input, balanced (XLR) analog outputs, with γ3's up-sample rate set to 192KHz, anti-clip mode disabled, non-inverting absolute phase, and analog output load impedance of 100KΩ. Other sample rates, input media (AES/EBU, optical or USB) and unbalanced analog output produced similar results.

Analog output impedance
(1KHz)
66Ω balanced
33Ω unbalanced
Frequency response See graphs below
S/N ratio
(RMS, A-weighted, digital zero, relative to 0dBFS)
121.6dB (minimum volume)
118.0dB (maximum volume)
Noise floor
(RMS, unweighted, digital zero, relative to 0dBFS, at 1KHz)
-152.0dB (minimum volume)
-149.8dB (maximum volume)
Total harmonic distortion + noise
(A-weighted, 1KHz at -1dBFS, volume set to maximum)
0.00060%
Intermodulation distortion + noise
(A-weighted, SMPTE 60Hz:7KHz 4:1, volume set to maximum)
0.00073%
Stereo crosstalk
(0dBFS, volume set to maximum)
-125.5dB at 1KHz
-113.0dB at 20KHz
Jitter
(J-Test method)
unmeasurable
Anti-clip mode behavior 2dB attenuation
AC power consumption 4W standby
20W maximum active

Performance graphs

The graphs below are as measured on the Audio Precision audio analyzer. For a "loop-back" baseline of the analyzer's measurement limits, click here.

Click on a graph to see it in full size.

Frequency response (24b/192KHz)

The following graphs shows the frequency response of each of the five user-selectable filters.

Resp 1: Linear Phase "soft knee" filter
Resp 2: Minimum Phase "soft knee" filter
Resp 3: Linear Phase "brick wall" filter
Resp 4: Minimum Phase apodizing filter
Resp 5: Linear Phase apodizing filter
Frequency response (24b/192KHz)

Here is the same data as before, but zoomed in for more detail.
 
Frequency response (filter 3)

The frequency response sweep with filter response 3, but with 44.1KHz, 48KHz, 96KHz and 192KHz sample rate data, each sweep is up to fs/2:
Frequency response (filter 3)

Here is the same data as before, but zoomed in for more detail.
 
Noise floor (digital zero, minimum volume)

The unweighted FFT noise floor spectrum with the volume control set to minimum. Much of the noise floor lies around -150dB or slightly below.
Noise floor (digital zero, maximum volume)

The unweighted FFT noise floor spectrum with the volume control set to maximum. It is only slightly higher than at minimum volume.
 
THD+N, 1KHz sine wave (0dBu)

FFT spectrum showing the fundamental 1KHz tone and its harmonics. The harmonics are very low in level. The highest spike is the 5th harmonic at -115dB, corresponding to 0.000178%. The next highest is the second harmonc at -116dB, or 0.000158%. All other harmoncs are -120dB (0.0001%) or less.
THD+N, 1KHz sine wave (0dBu, 16b/44.1KHz)

Similar to the 24b/44.1KHz graph, but with a 16b/44.1KHz tone, the harmonics are just slightly higher but mostly buried in noise (due to less resolution and about 8dB increase in noise).
 
IMD+N (SMPTE)

The unweighted FFT spectrum of the SMPTE intermodulation distortion test tones. The spikes at 120Hz, 180Hz and 240Hz are actually harmonics of the 60Hz tone, not intermodulation distortion.
Stereo crosstalk vs. frequency

Here is a stereo crosstalk vs. frequency graph. There is a general up-tilt as the frequency increases, but even at 20KHz the crosstalk is no more than -113dB.
 
THD+N vs. frequency

The graph below shows the γ3's unweighted RMS THD+N vs. frequency. Since no weighting is applied, the distortion level is about 0.0002% higher than the A-weighted result shown previously, and basically flat across the entire audio band. The downward tilt at high frequencies is a measurement artifact (due to the 22KHz LPF in the Audio Precision analyzer, enabled to prevent out-of-band artifacts from affecting the results).
D/A amplitude linearity (24b/48KHz)

An interesting test of the DAC's amplitude linearity. The yellow trace (right side scale) shows the input vs. output amplitude, with the horizontal scale expressed in terms of bits of resolution. A perfect DAC would have a straight line from upper right to lower left. The γ3 is essentially perfect down to 20.5 bits, which is excellent performance. The cyan trace shows the deviation from linearity (left side scale).
 
Low-level resolution (16b/44.1KHz)

An undithered 16 bit, 44.1KHz sine wave tone at 1KHz, with an amplitude of -90.31dBFS (a very faint signal) is fed into the γ3, and the resultant output waveform is shown below. At this low level, with 16 bits of resolution, the sine wave can only be represented by 3 DC voltage level "steps". The γ3 produced al almost picture-perfect version of this waveform, the ringing at each level is noise.
Low-level resolution (24b/44.1KHz)

With 24 bit data at the same -90.31dB amplitude, the output is a good, low-distortion representation of the sine wave with the noise floor superimposed on it.
 
Low-level resolution FFT spectrum (16b/44.1KHz)

The following is the FFT spectragram showing a dithered 16 bit, 1KHz, -90dBFS tone output. Note that all distortion products are completely buried under the noise floor.
Low-level resolution FFT spectrum (24b/44.1KHz)

Changing to dithered 24 bit data, the noise floor drops by around 26dB compared to the 16 bit data. This is superb performance, and corroborates with the 20.5 bits of D/A linearity shown previously.
 
Jitter rejection (J-Test, 16b/44.1KHz)

J-Test is a high level 11.025KHz sine wave combined with a low level 229.6875Hz square wave. When a 16-bit version of this signal is decoded by the γ3, the result is the following spectrum. The evenly-spaced spikes in the spectrum are a part of the J-Test signal itself, as is the large "carrier" tone at 11.025KHz. Jitter-related artifacts will show up as other symmetrical side-bands around the carrier. As you can see, there are no such side-bands. In other words, the jitter is so low that it is practically zero. Also note that the J-Test spikes are each at their correct levels, described by the downward-tilting red line in the graph.
Jitter rejection (J-Test, 16b/44.1KHz)

Here is the same data, but zoomed in along the horizontal axis. Again, no jitter-related sidebands are visible. The combined benefits of ASRC, low-jitter oscillator, and other γ3 design elements paid off handsomely in these test results.
 
Jitter rejection (J-Test, 24b/44.1KHz)

With 24 bit J-Test data, the output is just a clean central carrier spike without other artifacts.
Frequency response with pre-emphasis

The following is a frequency response sweep with 16 bit, 44.1KHz data at -12dBFS, and with 50/15µs pre-emphasis enabled. The cyan trace shows the resultant flat frequency response with de-emphasis set to Auto. The green trace shows the boosted treble when de-emphasis is set to Off.

Oscilloscope waveforms

Some γ3 measurements were made with the Tektronix TDS2014B 100MHz quad trace digital storage oscilloscope. Input waveforms were produced either by the Altor Audio JK-GEN 384 digital signal generator or by software on the USB host computer. Sine wave and triangle wave response were textbook-perfect, so they are not shown. Only square wave response and pulse response are shown below.

Square wave response

The γ3's 100Hz, 1KHz and 10KHz square wave response with 44.1KHz and 192KHz sample rates are shown below. The up-sample rate was set to 192KHz and filter C (linear phase brick wall) filter was selected for these measurements. We can clearly see how a higher sample rate (and wider bandwidth) improves the square wave. The ringing in the leading and falling edges are due to Gibbs Phenomenon. With low sample rates such as 44.1KHz or 48KHz, the 10KHz square wave turned into a sine wave because the filter removed all ultrasonic harmonics just beyond audio band.

44.1KHz sample rate 192KHz sample rate
100Hz square wave
100Hz square wave
1KHz square wave
1KHz square wave
10KHz square wave
10KHz square wave

Pulse response

These oscillograms show the output when the input is a 40µs wide pulse at 192KHz sample rate. We examine the difference between 96KHz and 192KHz up-sample frequency settings. As can be seen, the Gibbs Phenomenon ringing characteristics change with the selected filter.

The linear phase filters all exhibit symmetrical ringing before and after the pulse, whereas the minimum phase filters have no pre-ringing. The linear phase soft knee and apodizing filters have less pronounced ringing than the brick wall filter.

96KHz up-sample rate 192KHz up-sample rate
filter 1 (linear phase soft knee)
filter 1 (linear phase soft knee)
filter 2 (minimum phase soft knee)
filter 2 (minimum phase soft knee)
filter 3 (linear phase brick wall)
filter 3 (linear phase brick wall)
filter 4 (minimum phase apodizing)
filter 4 (minimum phase apodizing)
filter 5 (linear phase apodizing)
filter 5 (linear phase apodizing)



Main: γ3 Main | Prev: History & acknowledgements | Next: Buy