Well, before measuring let's discuss listening test results.
Zero oversampling, as said, has one real issue: a drop in the frequency response increasing with frequency and reaching a few dBs at 20KHz.
What's worst, there is no "technically perfect" way of solving this issue. The best one, from the technical standpoint, is still oversampling, but the nasties introduced by this at sonic level show that also this solution is far from being globally perfect.
Another approach is to use a corrective analog filter. Unfortunately any non-digital filter causes a phase rotation that, according to most of the people who tested this way, produces sonic effects even worse than oversampling.
So, after finding this totally unexpected frequency response, my reasoning was this: "The frequency response is completely different from any zero oversampling DAC I have ever listened to; apparently, there must be an output filter correcting the drop. However, I did not think Fabio could choose such a rough solution.".
I was wrong, but I was right.
First measure, out of band noise at the audio output while playing a sinusoid slowly sweeping through audio band.
The general format, with huge lobes even after 22KHz, is typical of zero oversampling, and there is no way to avoid it: any zero oversampling DAC has these lobes, even though they can be attenuated by filters. However there is nothing that shows the presence of any kind of output filter. Strange.
Second measure, frequency response from 10Hz to 22KHz.
Anything strange here? No... or better yes again. The frequency response is dropping at high frequencies slightly more than the TNT1541: the measurement does not match at all with listening experience, the difference between the balance shown by measurements and the one perceived in listening is definitely very relevant!
What's going on here, then???
Let's try to measure jitter.
OK, already the first look tells us that there is something really strange. I used the usual jitter test signal, a -6dB, 11.024 kHz (Fc/4) pure tone with the LSB bit toggling every 192 samples: as you see here the FFT is presented using a full scale diagram (normally I present the -130:-100dB range), and the high noise floor and the multiple lines around the center tone are quite evident even at this large scale, if compared with the reference ones. The figures confirm the situation, with a jitter value oscillating between 3.9ns e 5.9ns. Such a value cannot be obtained casually, it is too high (the worst case I measured before this was 1.5nsec), and in particular Fabio, who has been distributing for years the LC Clocks in Italy is too expert to make such a disaster. So what's going on???
There is only one explanation that is compatible with all the problem terms.
Asynchronous reclocking is another hot topic in DIY hi end audio circles together with zero oversampling. The inventor once again is possibly Mr.Kusunoki, who presented a circuit applying this technique and known as "Improved Simple DAC." in the December 1997 issue of MJ.
The operation in practice is very simple: after the digital receiver, you just "resample" all the lines from the receiver to the DAC with a high frequency, very stable and low jitter clock, at 50 or 100MHz. The signals in the lines are far slower than this clock, so that this modification does not change significantly the global DAC behaviour. However, it does affect the conversion timing, because one of the lines is the LRCK, that usually directly drives the conversion.
According to Kusunoki, "The operation of the system is quite simple: to rectify the clock signal generated by the DAI-IC [DAI = Digital Audio Interface], such as a CS8412, with another free-running clock. The problem here is the reason why we feel the expansion of sound stage? Other friends and myself verified the existence of this phenomenon. My presumption is that the clock signal generated by the existing DAIs contain jitter that correlate to the source signal, and the re-clocking breaks up this correlation."
In facts, the low level lines at multiples of 229Hz present in the test signal just to introduce a pattern correlated with the SPDIF flow, and visible in the TNT1541 spectrum, here are completely hidden in the noise floor, and there is no corresponding relevant line in DC-1 spectrum.
As a matter of fact, some aspects of the sound of DC-1 make it in some extent similar to the Weiss Medea: in particular the apparently very detailed high frequencies and the overall rather open balance.
However, from a purely technical standpoint, the situation is far different. Normally, the conversion clock has a period of exactly 1/44100s = 22.6757us. When the conversion clock is reclocked with a 50MHz clock, in each conversion clock period 1133.787 cycles of the 50MHz clock take place. This means that each conversion cycle will assume one of two possible different values, given by the two integer number nearest to 1133.787, that is 1133 and 1134, times the 50MHz clock cycles, that is 22.66 or 22.68us.
In practice, we have introduced a peak-to-peak jitter of 20nsec. The corresponding RMS jitter is lower, around 7-8ns based on very rough calculations, which is not so far from the measured values.
Well, who blamed me for not explaining clearly which is jitter sound, now can understand why I had to be so reticent. Jitter has many different characters, is terribly elusive, because it is not so important the amount of jitter, but its quality; and the definition of quality, for jitter, is extremely complex, as is for any type of noise. Frequency and statistical behaviour are just partial representation, probably far from enough to completely describe its effect on the human ear.
Moreover, the effect of clock jitter on DAC chips utterly depends on DAC characteristics, so that the global effect is far from being easily foreseeable or even measurable.
One way to reduce its effects is using conversion chips less affected by clock jitter: zero oversampling, especially with r-2r DACs, seems in facts less affected, for example, and this can give a partial explanation why the very large jitter in DC-1 is not perceived at all as such.
The only way to be sure to have no effect at all is to eliminate it completely, but this is very expensive.
Another way is to simply try to use jitter to get the desired sound. Because of its rather unpredictable effects, this is hardly possible, but when a specific kind of jitter has a specific and reproducible effect, then I cannot see any good reason to avoid it.
Not Hi Fi??? I do not know many people that say that zero oversampling is not HiFi; well, please, have a look at the frequency response below: they probably do not even match the ancient and forgotten DIN hi-fi norms. And we call HiFi something with such a frequency response... Single driver speakers have even worse behaviours, in many cases: and we accept them.
The problem is that Hi-Fi is all about trade-offs: you just cannot get a flat frequency response and the natural sound of zero oversampling at the same time, neither the perfect time alignment of single drivers with an extended frequency response.
So Fabio has just decided to use one of the existing possible solutions to solve the very evident frequency response issue.
It is not a perfect solution? Sure, but what's perfect in our world?
All that I can really say, based on listening tests, is that I had no perception at all of any clock jitter, and I gathered the only evidence in measuring it. Even afterwords, I tried to understand if any of the jitter related sound issues normally apparent when listening to low cost players was perceivable, but pace and rhythms are simply very good, and music always makes perfect sense.
Back to [Part 1] | Go to [Part 3]
© Copyright 2005 Giorgio Pozzoli - www.tnt-audio.com