Head-Direct HiFiMan HM-602 Digital Audio Player Measurements
To check out the performance of this $439 player on the test bench, I used Stereophile's loan sample of the top-of-the-line Audio Precision SYS2722 system (see www.ap.com and the January 2008 "As We See It"). For some tests, I also used my Audio Precision System One Dual Domain and the Miller Audio Research Jitter Analyzer. To test the HiFiMan's performance as a USB DAC, I used my MacBook running OS10.6 and Bias Peak Pro 6 to play WAV files, switching the HM-602 from "DAP" to "USB" mode with its top-panel switch. To test it in DAP mode, I loaded WAV and FLAC files representing 16- and 24-bit test tonesthe HM-602 doesn't play AIFF or ALAC filesinto the HiFiMan's internal flash memory, a procedure as simple as plugging the player into my computer's USB port and dragging'n'dropping the files.
In USB mode, Apple's USB Prober utility identified the HM-602 as being a two-channel "USB PnP Sound Device" manufactured by "C-Media Electronics Inc." and operating in isochronous adaptive mode. The bit depth was identified as 16, and the HM-602's USB receiver operated at just two frequencies: 44.1 and 48kHz. By contrast, in DAP mode, in which the HM-602 can access data stored in its internal memory, it played WAV and MP3 files with sample rates up to 48kHz and FLAC files up to 96kHz, including 88.2kHz. An oddity was that while the HM-602 would play 24-bit FLAC files, trying to play a 24-bit WAV file caused the player to momentarily freeze, then reboot. The measurements were taken with the HM-602's battery fully charged and its equalization turned off. Turning the EQ on and setting it to Normal didn't change the measured performance.
The maximum output level from the line-level output was 1.79V, sourced from a 214 ohm impedance at high and middle frequencies. This increased to 558 ohms at 20Hz. In the Low Gain setting, the headphone output's maximum level was 1.85V, sourced from a constant 12 ohms across the audioband; High Gain increased the maximum level by 3.2dB, to 2.69V. While the headphone output preserved absolute polarity, the line output inverted polarity, as can be seen in fig.1, which shows the HM-602's impulse response. More interesting, this response indicates that the HM-602 either doesn't use a digital reconstruction filter, or uses one optimized for its time-domain behavior. Head-Direct's Fang Bian confirmed that the HiFiMan player doesn't use a digital reconstruction filter, just a slow-rolloff analog low-pass filter. While exotic D/A processors from Audio Note and Zanden use the so called "non-oversampling" topology, the HM-602 appears to be unique in that it is the first portable player to go NOS.
Fig.1 HiFiMan HM-602, response to single sample at 0dBFS, 44.1kHz-sampled data (4ms time window).
Fig.2 shows the HM-602's frequency response with 44.1kHz data. While the low-frequency rolloff is very mild, the top-octave output does roll off prematurely, due to the analog low-pass filter that follows the DAC. This should not in itself have any audible consequences, however. The channel matching is good, with just a 0.04dB imbalance evident. Channel separation (not shown) was good rather than great, at 90dB in the midband, decreasing slightly at the frequency extremes.
Fig.2 HiFiMan HM-602, frequency response at 12dBFS into 100k ohms from line outputs with data sampled at 44.1kHz (left channel blue, right red; 0.25dB/vertical div.).
The TDA1543 D/A chip used by the HiFiMan dates back to the early 1990s, and is a 16-bit part. So while the HM-602 will play 24-bit data files, and while doing so even displays "24bit" rather than "16bit" on its screen, it will always truncate those data to 16 bits when they are presented to the DAC.
The TDA1543 does operate up to 192kHz but doesn't offer the excellent low-level performance offered by more modern chips. Fig.3, for example, which shows a 1/3-octave analysis of the player's line output while it played a 16-bit WAV file representing a dithered 1kHz tone at 90dBFS, indicates that both channels suffer from a high level of second-harmonic distortion, and that the left channel (solid trace) also has a significant degree of negative amplitude error.
Fig.3 HiFiMan HM-602, 1/3-octave spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS with 16-bit data (right channel dashed).
This was confirmed by FFT analysis (fig.4) and by the plot of linearity error (fig.5). The top trace in fig.5 is the right channel; as the tone drops below 85dBFS, the measurement begins to be affected by noise. By contrast, the left channel (bottom trace) suffers from an increasing amount of negative amplitude error as it drops below 70dBFS. What is happening is that the DAC has some missing codes and, as a result, is starting to act as a frequency doubler at low levels, pumping energy into a tone an octave higher (hence the high level of second-harmonic distortion) and, as it does, reducing the level of the fundamental. You can just make this out in the plot of an undithered waveform at 90.31dBFS (fig.6), where the left channel (blue trace) has more zero crossings than the right (red), though this is obscured by the high level of high-frequency noise.
Fig.4 HiFiMan HM-602, FFT-derived spectrum with noise and spuriae of dithered 1kHz tone at 90dBFS with 16-bit data (left channel blue, right red).
Fig.5 HiFiMan HM-602, linearity error of left (bottom) and right (top) channels, 16-bit data (2dB/vertical div.).
Fig.6 HiFiMan HM-602, waveform of undithered 1kHz sinewave at 90.31dBFS, 16-bit data (left channel blue, right red).
The HM-602's distortion was generally higher than I like to see, with both many harmonics evident between 80 and 90dB (0.0030.01%), and a granular-looking noise floor with the headphone output set to Low Gain and driving a full-scale tone into 300 ohms (fig.7). The line output was only slightly better driving 100k ohms, and couldn't drive loads below 1k ohm at full level without asymmetrical waveform clipping appearing. Don't plug your headphones into the line output!
Fig.7 HiFiMan HM-602, headphone output, Low Gain, spectrum of 1kHz sinewave, DC10kHz, at 0dBFS into 300 ohms (left channel blue, right red; linear frequency scale).
The granular noise floor can also be seen in the spectrum of the line output while the player decoded a 16-bit WAV file representing an equal mix of 19 and 20kHz tones (fig.8). More important, the absence of a reconstruction filter gives rise to very strong images of the tones appearing at 24.1 and 25.1kHz, as well as other enharmonic products that fold down into the audioband. This transferal of ultrasonic energy into the audioband was also evident with 96kHz-sampled data.
Fig.8 HiFiMan HM-602, HF intermodulation spectrum, DC24kHz, 19+20kHz at 0dBFS into 100k ohms (linear frequency scale).
Playing 16-bit data from its internal memory, the HM-602 offered quite good rejection of jitter. The spectrum of its output while playing the J-Test tone (fig.9) does show some emphasis of the 229Hz-spaced sidebands, which the Miller Analyzer calculated at 659.4 picoseconds peakpeak in the left channel and 573ps pp in the right. Operated as a USB DAC, however, the HM-602 gave a level of jitter so high that the Miller Analyzer couldn't reliably measure it. The spectrum is shown in fig.10: Not only are the 229Hz-spaced sidebands extremely high in level; the central peak that represents the 11.025kHz tone also has a significantly broadened base, suggesting the presence of very high random low-frequency timing variations. This poor performance rules the HM-602 out of contention for use as a USB DAC, in my opinion, but it's possible that our sample was broken.
Fig.9 HiFiMan HM-602, high-resolution jitter spectrum of analog output signal, 11.025kHz at 6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit internal data (left channel blue, right red). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Fig.10 HiFiMan HM-602, high-resolution jitter spectrum of analog output signal, 11.025kHz at 6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit data via USB (left channel blue, right red). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
In playing 16-bit/44.1kHz music files from its internal memory, the Head-Direct HiFiMan HM-602's sound will be dominated by the designer's decision to forgo a reconstruction filter. There are those who strongly believe that such an approach results in digital sound quality that is considerably more analog-like. But I still don't much like the behavior of the TDA1543 D/A chipmy iPod Classic 160GB offers significantly better measured performance than the HM-602 in every way.
There was much discussion on the Internet about the fact that the HM-602 will play high-resolution files. But, as I've pointed out here, it truncates 24-bit files to 16 bits, and its NOS DAC topology doesn't measure well with high sample rates; I'm not sure, therefore, what benefit other than convenience this ability confers on the HM-602.John Atkinson