Quote from Pedja on 3 December 2021, 12:44 PM

It is more than half a year since we started shipping the S5 DAC, and as I promised to some of you, here is the public reply regarding some news brought by the S5 design.

The first is about the TDA1541A DEM circuit clocking: as you know, the S5 abandoned synchronous DEM clocking. Also, the latest AYA 5 DAC project includes the required circuit on the board but does not actually include it in the typical configuration, leaving it as an optional feature instead. So, why did it turn like that?

First, about the DEM (Dynamic Element Matching) itself. It denotes the circuit which interchanges the currents in time, in order to produce the equal resulting currents. Several articles written by Philips engineers back in the ’70s and ’80s describe such an approach.

Figure 2 you will find in the Schouwenaars, Dijkmans, Kup, and van Tuijl's "A Monolithic Dual 16-Bit D/A Converter" can not be more informative on this. Please note the bottom right part of the drawing, marked by the rectangle. This way, four possibly unequal current sources outputs, interchanged in time, produce four equal currents.

 

 

Of course, such a circuit requires some oscillator to work. For this purpose, the TDA1541A includes a bistable multivibrator, which frequency is controlled by the external capacitor. (Original TDA1541, non-A, employed this capacitor inside.) The frequency generally recommended by Philips is between 150 kHz and 300 kHz, with the bottom limit at 80 kHz (it is internally divided by four, as also implied by the above picture, and this divided-by-four frequency should not come inside the audio band), while above 300 kHz the performance slightly deteriorates (very slightly though, and mostly at the bottom end).

 

 

Typically used 470 pF capacitor sets this frequency to about 200 kHz. And this is the typical setup, that makes a "free-running" oscillator, meaning that it is not in any way associated with the digital audio stream clocking.

The same capacitor nodes can be however also used to connect the external source for this oscillator frequency. And I was using the external circuits to clock the TDA1541A synchronously with the audio stream, for more than 10 years. Now, why this "step back"?

The first reason, chronologically speaking, was practical, and associated with the move from CS8414 to CS8416 S/PDIF receiver. Some will say "a somewhat late move", but to my defense, I actually used CS8416 years ago, in D-09, where I needed up to 192 kHz capable S/PDIF receiver. With TDA1541A, this really was no requirement, and since CS8416 brought no major performance improvements (besides its ability to lock on preamble), I just stuck with CS8414, so long as it was possible. Now, since the CS8414 became somewhat tricky to source, and I also wanted more inputs, I found CS8416 again useful.

Now, what did this change about the synchronous DEM clocking? The CS8416, with its PLL unlocked, outputs lower (bit and word) clock frequencies than the CS8414. Since my synchronous DEM clock is tied to the bit clock, and the DEM clock is, as said above, also further divided internally by four by the TDA1541A, its final frequency falls notably inside the audio band, and DEM becomes "audible". Practically speaking, when you select BNC or optical input and do not connect anything there, you will experience a sort of noise. Although it is relatively low in level, and it stops once the audio stream starts, such things give the impression of a "poor design".

CS8416 offers also a way out of this, but it requires one more oscillator, and I was reluctant about that, as there are already several HF sources inside, and I paid a lot of attention in order to put them all under control.

So, many years later, and taking into account the S5 as an environment, I switched this back (and then forth, and then back), to revise the sonic effects of synchronous DEM clocking.

And, synchronous and a classic DEM clocking definitely draw somewhat different pictures. Synchronous clocking brings more sharp edges and positioning. The bass gets not only more tightened, but it also feels better tuned, in a way it never sounds like one-note. I especially liked this in a time when opinions on the muddy or woolly bass performance of TDA1541A practically prevailed. (I knew that such fame rather came from many other parts of TDA1541A circuit designs I've seen - all the way from the early Philips CDs through the "advanced" output stage designs with tubes - but, you know...)

My later designs, according to the customers' feedback, also brought more resolution, in a way that really does not leave anything to be missed. Removing synchronous and applying again free-running clocking for DEM, at this time showed some good sides of it: a bigger scale, more ambiance, and more velvet tone. Cymbals are interesting, as they are probably not that sparkling like with synchronous clocking, but are actually cleaner, and also have more body at that. It is, maybe not entirely, but a sort of Paiste vs Zildjian case. And although they tonally act a bit differently, both these options, in the current circumstances, have the same or similar overall definition.

Finally, my decision for S5 was to proceed with a free-running clock for DEM. The AYA 5, which is partially intended for DIY endeavors and systems, where things can be individually tweaked, leaves synchronous clocking as an optional feature.