Denafrips Terminator-Plus Digital-to-Analog Converter Measurements

Link: reviewed by Evan McCosham on SoundStage! Hi-Fi on September 1, 2021

General Information

All measurements taken using an Audio Precision APx555 B Series analyzer.

The Terminator-Plus was conditioned for 30 minutes at 0dBFS (4.3Vrms out) into 200k ohms before any measurements were taken.

The Terminator-Plus offers several digital inputs: AES-EBU (XLR), coaxial S/PDIF (RCA and BNC), optical S/PDIF (TosLink), and USB. There are two line-level outputs (balanced XLR and unbalanced RCA). Comparisons were made between unbalanced and balanced line-level outputs, and aside from the 6dB extra voltage over balanced, there were no appreciable differences observed in terms of THD and noise when the output voltage was normalized. Noise was slightly lower over the balanced outputs, yielding a little more than the expected 6dB boost in dynamic range over the unbalanced outputs (129 vs 121dB  with 24/192 data, A-weighted).

There are two filter settings, Sharp and Slow, available in OS (oversampling) mode. There is also a NOS (non-oversampling) mode that can be activated. All measurements below, unless otherwise stated, are for the coaxial digital input, and the balanced output, using the Sharp filter.

Despite Denafrips’ claim that the DAC can be operated in NOS mode, according to our measurements and the explanations provided by the GoldenSound YouTube Channel, it appears that NOS mode is not actually true NOS for the following reasons.

• When wideband (1MHz) FFTs are observed at 44.1kHz and 96kHz sample rates (see graphs further in the report), there is a peak in the spectrum at 352.8kHz (exactly 8x oversampling) in the 44.1kHz spectrum and 384kHz (exactly 4x oversampling) in the 96kHz spectrum, for both NOS and OS modes. No oversampling peaks should be seen.
  
  
• In a NOS DAC, square-wave rise times should be the same regardless of sample rate, however, with the Terminator-Plus in NOS mode, we found the rise time with a 100Hz square wave to be exactly 2x as fast when sampled at 96kHz (12.5us) versus 48kHz (25us).
  
  
• The reproduction of a 15kHz sinewave (not published here) in NOS mode appeared to have linear interpolation applied.
  
  
• The theoretical attenuation for a sine wave at 22.05kHz for a true NOS DAC is -3.9dB, while we measured -4.9dB.
  
  
• Finally, the impulse response of a NOS DAC should be a square, however, the Terminator-Plus in NOS mode exhibited a triangular-shaped impulse response (see graph further below), indicating once again, some form of linear interpolation between samples.

Published specifications vs. our primary measurements

The table below summarizes the measurements published by Denafrips for the Terminator-Plus compared directly against our own. The published specifications are sourced from Denafrips’ website, either directly or from the manual available for download, or a combination thereof. With the exception of frequency response, where the Audio Precision bandwidth is set at its maximum (DC to 1MHz), assume, unless otherwise stated, the coaxial digital input (24-bit/96kHz 1kHz sine wave at 0dBFS), the balanced line-level outputs into 200k ohms using a measurement input bandwidth of 10Hz to 90kHz, and the worst-case measured result between the left and right channel.

Our primary measurements revealed the following using the coaxial input and the balanced line-level outputs (unless specified, assume a 1kHz sine wave at 0dBFS, 200k ohms loading, 10Hz to 90kHz bandwidth, OS with Sharp filter engaged):

Frequency response (16/44.1, 24/96, 24/192, all OS mode with Sharp filter)

The plot above shows the Terminator-Plus frequency response as a function of sample rate. The blue/red traces are for a 16-bit/44.1kHz dithered digital input signal from 5Hz to 22kHz, the purple/green traces are for a 24/96 dithered digital input signal from 5Hz to 48kHz, and finally orange/pink represents 24/192 from 5Hz to 96kHz. The behavior at low frequencies is the same for the digital input types—perfectly flat down to 5Hz. The behavior at high frequencies for all three digital sample rates is as expected, offering sharp filtering around 22k, 48k, and 96kHz (half the respective sample rates for each). The -3dB point for each sample rate is roughly 21.2, 47, and 84kHz, respectively. It is also obvious from the plots above that the 44.1kHz sampled input signal offers the most “brick-wall”- type behavior, while the attenuation of the 96kHz and 192kHz sampled input signals approaching the corner frequencies (48kHz and 96kHz) is more gentle. In the graph above and most of the graphs below, only a single trace may be visible. This is because the left channel (blue, purple, or orange trace) is performing identically to the right channel (red, green, or pink trace), and so they perfectly overlap, indicating that the two channels are ideally matched.

Frequency response (16/44.1 with Sharp and Slow filters, and NOS mode)

The plots above show frequency response for a 0dBFS input signal sampled at 44.1kHz for the Sharp filter (blue), the Slow filter (purple), and NOS mode (orange) into a 200k ohm-load for the left channel only. The graph is zoomed in from 1kHz to 22kHz, and extra sampling points were introduced around the corner frequency (the “knee”) to highlight the various responses of the three filters. We can see that the Sharp filter provides the most “brick-wall”-type response, while the Slow filter shows earlier attenuation but is similar to the Sharp filter. In NOS mode, the behavior is similar to a single-pole analog low-pass filter, with a corner frequency at roughly 15kHz. The -3dB points are at 21.5kHz, 20.9kHz, and 14.5kHz for Sharp, Slow, and NOS respectively. 

Phase response vs. sample rate (16/44.1, 24/96, 24/192, all OS mode with Sharp filter)

Above are the phase response plots from 20Hz to 20kHz for the coaxial input, measured at the balanced output, using the Sharp filter setting. The blue/red traces are for a dithered 16/44.1 input signal at -20dBFS, the purple/green for 24/96, and the orange/pink for 24/192. Additionally the brown/green traces are for the 24/96 data but with the Phase button disengaged, showing that the Phase function works as advertised (180 degrees shift, though the Phase button must be engaged for correct polarity). There’s a worst-case phase shift of around 80 degrees at 20kHz for the 16/44.1 signals, and less than 20 degrees for the 24/96 and 24/192 signals.

Digital linearity (16/44.1 and 24/96 data)

The chart above shows the results of a linearity test for the coaxial digital input (the optical input performed identically) for both 16/44.1 (blue/red) and 24/96 (purple/green) input data, measured at the balanced line-level output. The digital input is swept with a dithered 1kHz input signal from -120dBFS to 0dBFS, and the output is analyzed by the APx555. The ideal response is a straight flat line at 0dB (i.e., the amplitude of the digital input perfectly matches the amplitude of the measured analog output). The 24/96 input data showed a worst-case -1dB deviation just above -120dBFS, and the 16/44.1 data +3dB at -120dBFS. Linearity was also verified below -120dBFS down to -140dBFS, where both 24/96 and 16/44.1 data deviated considerably from the 0dB reference.

Impulse response using Audio Precision Transfer Function measurement (Sharp and Slow filters, and NOS mode)

The chart above shows the impulse response for the two different filter types in OS and NOS modes, for a -20dBFS 16/44.1 dithered input stimulus, measured at the balanced line-level output. The blue plot represents the Sharp filter, the purple represent the Slow filter, and the orange represents NOS mode. We see virtually identical symmetry and behavior in the Sharp and Slow filter settings. NOS mode also yielded a symmetrical impulse response, but with less pre- and post-ringing effects. Of note here is that this is not the shape of NOS DAC impulse response (it should be a square), but that of a typical sinc filter.

Impulse Response using Audio Precision Signal Acquisition measurement with true 0dBFS digital impulse WAV file (Sharp and Slow filters, and NOS mode)

The chart above shows the impulse response for the Terminator-Plus in NOS (blue/red) and OS (Sharp filter, purple/green) modes, for a looped 24/44.1 test file that moves from digital silence to full 0dBFS (all “1”s) for one sample period, then back to digital silence. The previous chart was generated using the Audio Precision Transfer Function measurement, which applies an inverse FFT to a noise signal applied to the DAC. We performed this measurement because we wanted to explore the NOS impulse response further with a true digital impulse applied to the input of the DAC. In OS mode, both this measurement and the previous one show the same typical sinc function response of the oversampling reconstruction filter. In NOS mode however, when zoomed in . . .

. . . we see a stair-stepped triangular-shaped impulse response, where, if the DAC where truly NOS, we would see a square shape. In addition, the frequency of the stair-stepping in the impulse response is exactly 352.8kHz, or 8x oversampling at 44.1kHz.

J-Test (coaxial input with Sharp filter)

The plot above shows the results of the J-Test test for the coaxial digital input measured at the balanced line-level output with the Sharp filter engaged. J-Test was developed by Julian Dunn the 1990s. It is a test signal—specifically, a -3dBFS undithered 12kHz square wave sampled (in this case) at 48kHz (24 bit). Since even the first odd harmonic (i.e., 36kHz) of the 12kHz square wave is removed by the bandwidth limitation of the sampling rate, we are left with a 12kHz sine wave (the main peak). In addition, an undithered 250Hz square wave at -144dBFS is mixed with the signal. This test file causes the 22 least significant bits to constantly toggle, which produces strong jitter spectral components at the 250Hz rate and its odd harmonics. The test file shows how susceptible the DAC and delivery interface are to jitter, which would manifest as peaks above the noise floor at 500Hz intervals (e.g., 250Hz, 750Hz, 1250Hz, etc.). Note that the alternating peaks are in the test file itself, but at levels of -144dBrA (fundamental at 250Hz) down to -170dBrA for the odd harmonics. The test file can also be used in conjunction with artificially injected sine-wave jitter by the Audio Precision, to also show how well the DAC rejects jitter.

We can see the primary low level peak at 250Hz at just below -140dBFS (which is in the test file), some of the subsequent harmonics below -150dBFS, and the worst-case peaks adjacent the main 12kHz peak at a very low -140dBrA. This outstanding J-Test FFT is an indication that the Terminator-Plus should not be sensitive to jitter.

J-Test (coaxial input in NOS mode)

The plot above shows the results of the J-Test test for the coaxial digital input measured at the balanced line-level output, in NOS mode. The FFT is similar but slightly worse that in normal OS (Sharp filter) mode, with the worst-case peaks adjacent the primary 12kHz peak reaching -135dBrA.

We also tried adding additional 2kHz sine-wave jitter using the APx555’s built-in jitter generator, in both OS and NOS modes, however, the Terminator-Plus has essentially perfect jitter immunity. Without any jitter immunity, clear sidebands peaks at 10kHz and 14kHz (12kHz main signal +/- 2kHz jitter signal) manifest in the FFT, however, with the Terminator-Plus, with both the coaxial and optical input (below) in OS and NOS modes, none could be seen even with 1000ns of added jitter. With the maximum allowable jitter by the APx555 (1592ns), the Terminator-Plus did however lose sync with the digital signal.

J-Test (optical input with Sharp filter)

The plot above shows the results of the J-Test test for the optical digital input measured at the balanced line-level output. For this test, the optical input yielded effectively the same, although slightly worse, results compared to the coaxial input, with a few more spurious peaks visible below -140dBrA.

J-Test (optical input in NOS mode)

The plot above shows the results of the J-Test test for the optical digital input measured at the balanced line-level output, in NOS mode. The FFT is basically identical to the NOS coaxial FFT plot above.

Wideband FFT spectrum of white noise and 19.1kHz sine-wave tone (Sharp filter)

The plot above shows a fast Fourier transform (FFT) of the balanced line-level output with white noise at -4dBFS (blue/red), and a 19.1 kHz sine wave at 0dBFS fed to the coaxial digital input, sampled at 16/44.1 (purple/green), using the Sharp filter setting. The fast roll-off above 20kHz in the white-noise spectrum shows the implementation of the brick-wall type reconstruction filter. There are only very small imaged aliasing artifacts in the audioband in the -125dBrA range. The primary aliasing signal at 25kHz is at -70dBrA, while the second, third, and fourth distortion harmonics (38.2, 57.3, 76.4kHz) of the 19.1kHz range from -95 to-120dBrA.

Wideband FFT spectrum of white noise and 19.1kHz sine-wave tone (Slow filter)

The plot above shows a fast Fourier transform (FFT) of the balanced-line level output with white noise at -4 dBFS (blue/red), and a 19.1 kHz sine wave at 0dBFS fed to the coaxial digital input, sampled at 16/44.1 (purple/green), using the Slow filter setting. The 19.1kHz FFT is full of distortion, and we are not sure why this occurred, but it was repeatable. Reducing the 19.1kHz signal down to -3dBFS resulted in a clean FFT, which we can see below . . .

Wideband FFT spectrum of white noise and 19.1kHz sine-wave tone (Slow filter, -3dBFS)

At -3dBFS, we can see a slightly slower roll-off above 20kHz in the white-noise spectrum for this Slow filter compared to the Fast filter. There is one more significant imaged aliasing artifact within the audioband at around 13kHz, measuring -115dBrA. The primary aliasing signal at 25kHz is significant at almost -25dBrA, while the second, third, and fourth distortion harmonics (38.2, 57.3, 76.4kHz) of the 19.1kHz tone are down near at -100dBrA or below. The IMD products between the sampling frequency (44.1kHz) and both the 19.1kHz primary tone and 25kHz alias at 63.2kHz and 69.1kHz are also predominant, at around -75dBrA.

Wideband FFT spectrum of white noise and 19.1kHz sine-wave tone (NOS mode)

The plot above shows a fast Fourier transform (FFT) at the balanced line-level output with white noise at -4dBFS (blue/red), and a 19.1 kHz sinewave at 0dBFS fed to the coaxial digital input, sampled at 16/44.1 (purple/green), using the NOS setting. We see the typical soft roll-off in the white-noise spectrum down to the 44.1kHz sampling frequency, then another rise and fall in the noise spectrum down to a doubling of the sampling frequency (88.2kHz), and so on. The worst-case imaged aliasing artifact in the audioband is at around 13kHz, measuring -105dBrA. Predictably, the primary aliasing signal at 25kHz is significant at -10dBrA, while the second, third, and fourth distortion harmonics (38.2, 57.3, 76.4kHz) of the 19.1kHz tone are in the -110dBrA range. The IMD products between the sampling frequency (44.1kHz) and both the 19.1kHz primary tone and 25kHz alias at 63.2kHz and 69.1kHz are also predominant at around -30dBrA.

THD ratio (unweighted) vs. frequency vs. load (24/96)

The chart above shows THD ratios at the balanced line-level output into 200k ohms (blue/red) and 600 ohms (purple/green) as a function of frequency for a 24/96 dithered 1kHz signal at the coaxial input. The 200k and 600 ohms data are very close to one another. Interestingly, the lower impedance load yielded the lower THD results. Between 20Hz and about 300Hz, there is a 5dB improvement in THD ratios at 600 ohms compared to 200k ohms, and also between right and left channels. Above 1kHz, the difference between all traces is no more than 2-3dB. THD values are as low as 0.0004% between 20 and 200Hz into 600 ohms for the right channel, and at worst, at little over 0.002% at 20kHz for the left channel into 200k ohms.

THD ratio (unweighted) vs. frequency vs. sample rate (16/44.1 and 24/96)

The chart above shows THD ratios at the balanced line-level output into 200k ohms as a function of frequency for a 16/44.1 (blue/red) and a 24/96 (purple/green) dithered 1kHz signal at the coaxial input. The 24/96 and 16/44.1 data are essentially identical, with the same trend of the right channel slightly outperforming the left channel. THD values are a little lower at lower frequencies, but overall, hover between 0.001 and 0.002% across the audioband.

THD ratio (unweighted) vs. output (16/44.1 and 24/96)

The chart above shows THD ratios measured at the balanced output as a function of output voltage for the coaxial input into 200k ohms from -90dBFS to 0dBFS at 16/44.1 (blue/red) and 24/96 (purple/green). The 24/96 outperformed the 16/44.1 data at lower input stimuli, with a THD range from 0.6% at 200uVrms down to 0.0004% at 3Vrms (right channel), while the 16/44.1 ranged from 5% at 200uVrms down to 0.0004% at 2Vrms (right channel).

THD+N ratio (unweighted) vs. output (16/44.1 and 24/96)

The chart above shows THD+N ratios measured at the balanced output as a function of output voltage for the coaxial input into 200k ohms from -90dBFS to 0dBFS at 16/44.1 (blue/red) and 24/96 (purple/green). Due to the lower noise floor at 24 bits, the 24/96 outperformed the 16/44.1 data throughout, with a THD+N range from 3% at 200uVrms down to just below 0.002% at 3Vrms, while the 16/44.1 varied from 50% at 200uVrms down to 0.003% at the maximum output voltage of 4.3Vrms.

FFT spectrum – 1kHz (digital input, 16/44.1 data at 0dBFS, with Sharp filter)

Shown above is the fast Fourier transform (FFT) for a 1kHz input sine-wave stimulus, measured at the balanced output into 200k ohm for the coaxial digital input, sampled at 16/44.1 in OS mode with the Sharp filter. The worst-case signal harmonic is at 3kHz near -95dBrA, or 0.002%, with other even and odd harmonics at -110dbRA, or 0.0003%, and below. There is a small peak due to power supply noise seen at the mains frequency of 60Hz at -125dBrA. We can also see a clear peak at 352.8kHz, indicating 8x oversampling (i.e., 44.1kHz x 8).

FFT spectrum – 1kHz (digital input, 16/44.1 data at 0dBFS, in NOS mode)

Shown above is the fast Fourier transform (FFT) for a 1kHz input sine-wave stimulus, measured at the balanced output into 200k ohm for the coaxial digital input, sampled at 16/44.1 in NOS mode. Up to 20kHz or so, the FFT is nearly identical to the one above taken in OS mode. Above 20kHz, however, it’s clear that far more noise artifacts can be seen, including the two predominant peaks at -65dBrA due to IMD products between the sampling frequency (44.1kHz) and signal (1kHz) at 43.1 and 45.1kHz. Here we can still see a clear peak at 352.8kHz, indicating that 8x oversampling is likely being applied.

FFT spectrum – 1kHz (digital input, 24/96 data at 0dBFS, with Sharp filter)

Shown above is the fast Fourier transform (FFT) for a 1kHz input sine-wave stimulus, measured at the balanced output into 200k ohms for the coaxial digital input, sampled at 24/96 in OS mode with the Sharp filter. Due to the increased bit depth, the noise floor is much lower compared to the 16/44.1 FFT. The signal harmonics are effectively the same as with the 16/44.1 FFT, with the worst-case peak at 3kHz near -95dBrA, or 0.002%. With the lower noise floor, we can not only see the 60Hz peak due to power supply noise at -125dBrA, but also a small peak at 180Hz (third harmonic) near -145dBrA. We can also see a clear peak at 384kHz, indicating 4x oversampling (i.e., 96kHz x 8).

FFT spectrum – 1kHz (digital input, 24/96 data at 0dBFS, in NOS mode)

Shown above is the fast Fourier transform (FFT) for a 1kHz input sine-wave stimulus, measured at the balanced output into 200k ohm for the coaxial digital input, sampled at 24/96 in NOS mode. Up to 20kHz or so, the FFT is nearly identical to the one above taken in OS mode. The FFT looks almost identical to the 24/96 FFT in OS mode above, with the exception of a predominant peak at -85dBrA due to the IMD product between the sampling frequency (96kHz) and signal (1kHz) at 95kHz. Here we can still see a clear peak at 384kHz, indicating that 4x oversampling is likely being applied.

FFT spectrum – 1kHz (digital input, 16/44.1 data at -90dBFS)

Shown above is the fast Fourier transform (FFT) for a 1kHz input sine-wave stimulus, measured at the balanced output into 200k ohm for the coaxial digital input, sampled at 16/44.1 at -90dBFS in OS mode using the Sharp filter. The signal peak has the correct amplitude, and we see no signal harmonics above the noise floor within the audioband, and the same power-supply noise peak at 60Hz seen in the 0dBFS FFTs.

FFT spectrum – 1kHz (digital input, 24/96 data at -90dBFS)

Shown above is the fast Fourier transform (FFT) for a 1kHz input sine-wave stimulus, measured at the balanced output into 200k ohm for the coaxial digital input, sampled at 24/96 at -90dBFS in OS mode using the Sharp filter. The signal peak has the correct amplitude, and we see several signal harmonic peaks within the audio-band just above and below a very low-140dBrA, or 0.00001%. The same -125dBrA 60Hz noise peak can be seen here again, along with several other noise related peaks, but all are near or below -140dBrA.

Intermodulation distortion FFT (18kHz + 19kHz summed stimulus, 16/44.1)

Shown above is an FFT of the intermodulation distortion (IMD) products for an 18kHz + 19kHz summed sine-wave stimulus tone measured at the balanced output into 200k ohms for the coaxial input at 16/44.1 in OS mode using the Sharp filter. The input dBFS values are set at -6.02dBFS so that, if summed for a mean frequency of 18.5kHz, would yield 4.3Vrms (0dBrA) at the output. We find that the second-order modulation product (i.e., the difference signal of 1kHz) is at -110/-125dBRA (right/left), or 0.0003/0.00006%, while the third-order modulation products, at 17kHz and 20kHz, are higher, at around -105dBrA, or 0.0006%.

Intermodulation distortion FFT (18kHz + 19kHz summed stimulus, 24/96)

Shown above is an FFT of the intermodulation distortion (IMD) products for an 18kHz + 19kHz summed sine-wave stimulus tone measured at the balanced output into 200k ohms for the coaxial input at 24/96 in OS mode using the Sharp filter. The input dBFS values are set at -6.02dBFS so that, if summed for a mean frequency of 18.5kHz, would yield 4.3Vrms (0dBrA) at the output. We find that the second-order modulation product (i.e., the difference signal of 1kHz) is at -110/-125dBRA (right/left), or 0.0003/0.00006%, while the third-order modulation products, at 17kHz and 20kHz, are higher, at around -105dBrA, or 0.0006%.

Diego Estan
Electronics Measurement Specialist