The inspiration for this project came from our review of the Marantz CD6003 CD player, which appeared in the June 2011 issue. At the time, we made some measurements using our Audio Precision System One and discovered that it not only had a very low harmonic distortion figure for a CD player but it was practically flat across the audible frequency band (20Hz-20kHz).
We figured that this was partly due to its Crystal (Cirrus Logic) CS4398 DAC (digital-to-analog converter) IC. This is mounted on a large PCB, amongst a forest of discrete and passive components. So we thought, hmmm . . . could we do something similar for our DAC design? We suspected they were also doing some fancy digital processing using a DSP (digital signal processor) to get that level of performance but that the CS4398 DAC must also be pretty good for such an excellent result.
It turns out we were right on both counts. The CS4398 is very good but Marantz seem to be doing some digital interpolation (possibly increasing the sampling rate to 96kHz or 192kHz) to keep the distortion so low. While our new DAC board does not have the benefit of digital interpolation, it is clearly superior to the previous design, especially when processing 24-bit/96kHz program material.
If you have already built a Stereo DAC kit and would like to try out this new board, it’s pretty easy. You just build the new PCB and swap it for the old one; it’s the same size and the critical parts are in the same locations. You then reprogram or swap the microcontroller on the input board and Bob’s your uncle.
Like the Marantz, we designed the filtering hardware using all discrete components (ie, bipolar transistors and passives).
There was some controversy on the internet (unheard of!) over our choice of op amps in the original DAC design (SILICON CHIP, September, October & November 2009). This time we have avoided using those “evil” little black boxes, which should make the extreme audiophile cognoscenti happy (impossible!).
The resulting circuit has a lot more components than it would if we had used op amps but they are all cheap and commonly available. The resulting wide bandwidth compared to an op amp means that the output filtering works very well.
Fig.1: harmonic distortion (ignoring noise) versus frequency for the original (DSD1796-based) and new (Crystal CS4398-based) DACs. The newer design has lower distortion overall but especially above 2kHz. The channels differ slightly due to layout asymmetries and differences in the ICs themselves. The spikes at 1.2kHz and 9kHz are due to aliasing between the test and sampling frequencies.
We tested both the original and new DAC designs extensively, using both our Audio Precision System One and the newer Audio Precision APx525 with digital processing. We also performed numerous listening tests, including blind A/B tests.
The first result that became clear from all this testing is that the original design really is very good. Its distortion and noise are low (including intermodulation distortion), its linearity is very good and it generally sounds excellent. However, the new DAC design measures even better, with lower distortion (especially at high frequencies), even lower intermodulation distortion and astounding linearity down to -100dB.
Fig.1 shows a comparison of the harmonic distortion between both channels of the original and the new DAC design. These tests were performed on the same unit with just the DAC boards swapped, so they give an apples-to-apples comparison.
Note that noise has been digitally filtered out of this measurement completely, for a couple of reasons. First, both DACs have quite a bit of high-frequency switching noise in their output (but a lot less than some DVD and Blu-ray players we’ve tested!) and this can mask the distortion if we set the bandwidth wide enough to capture harmonics of high audio frequencies. Second, the 20Hz-20kHz residual noise of both the original and new boards is similar and this too means that a THD+N comparison would tend to understate the reduction in harmonic distortion obtained with the newer design.
Fig.2: a comparison of channel separation (ie, crosstalk) for the original and new DAC boards. The original is slightly superior but both are very good, with less than -93dB crosstalk at any frequency and separation of at least 100dB up to 1kHz. As is typical, there’s more coupling in one direction (for the new design, left channel to right channel) than the other, again mainly due to asymmetry.
As you can see, harmonic distortion with the CS4398 is substantially lower than the original design, both at high frequencies (above 3kHz) and low frequencies (below 100Hz). The differences between channels are due to asymmetries in the PCB layout as well as mismatches between the two channels within the DAC ICs themselves (eg, due to resistor ladder tolerances).
Fig.2 shows the channel separation for both units. The lines labelled “left” show how much signal from the right channel couples into the left and the lines labelled “right” show the opposite. In both cases, channel separation is very good and is generally better than -100dB across the audio spectrum. The older design is slightly better in this respect, although the difference is largely academic.
Fig.3 compares the linearity of both DACs. This plot shows the deviation between the expected and actual output level for a sinewave at a range of levels between -60dB and -100dB. Both DACs perform extremely well in this test but the CS4398 is especially good, with a maximum deviation of no more than 0.25dB at -100dB! Its deviation is essentially zero above -84dB while the DSD1796 still shows some deviation up to -70dB.
Note that all of the above test results were obtained with the Audio Precision APx525 (which can test in the analog or digital domain) using 24-bit 96kHz signals fed into a TOSLINK input of the Stereo DAC project.
Fig.3: a comparison of the linearity of the original and updated DAC boards. Delta-Sigma DACs typically have good linearity and in fact both are excellent. However, the updated board (with the CS4398) is the best of the two with an astounding deviation of less than one quarter of a decibel at levels down to -100dB! (The dynamic range of CD-quality audio is just 96dB).
Fig.4 shows the FFT frequency spectra for the updated DAC with one channel in magenta and the other in khaki. This was computed with a one million sample window, an equi-ripple algorithm and 8x averaging. The test signal is at 1kHz and the bandwidth is 90kHz. The harmonics of the test signal are clearly visible at 2kHz, 3kHz, etc. Also visible is some 50Hz and 100Hz mains hum at around -120dB, as well as various intermodulation products of this hum with the fundamental and its harmonics.
As we said earlier, both DACs are very good but the updated design generally has better figures. We also ran the SMTPE intermodulation distortion test on both. This involves sending a 4:1 mix of 7kHz/400Hz sinewaves to the test device. These frequencies are then filtered from its output (400Hz with a high-pass filter and 7kHz with a notch filter) and the remaining harmonics measured. These will generally be the sum and difference frequencies of 6.6kHz and 7.4kHz but possibly other harmonics too.
The old design gives an intermodulation distortion level of around 0.0018% (-95dB) while the new design gives 0.0006% (-105dB); a significant improvement.