D  E  V  E  L  O  P  M  E  N  T        O  F        A        H  I  -  F  I
title
 
This is my own project of a Non Over Sampling Digital to Analog Converter using paralleled Philips TDA1543A, DIR9001 SPDIF receiver, passive I/V conversion and Direct Coupled output. I followed the advices of very experienced people that I personaly know and also used lots of information on the subject that are available online. It all started with Ryohei Kusunoki´s paper on the MJ japanese magazine, which was brought to my knowledge by my friends Marcelo Sfoggia (the famous recording and circuit's guru) and Márcio Chiaramonte, to whom I thank very much. I also thank André de Bacco for information on the fabrication of PCBs with laser printers and home iron, which by the way motivated me to give a course to my students showing the technique. On the Internet, I used most the information provided by Doede Douma, schematics and ideas from forums like us.hifidiy.net, www.diyaudio.com, www.diyhifi.org and many others. I also wish to thank Elso Kwak, Luigi Carro, and other friends for personal communication and comments about circuit details. Finally, I need to acknowledge my fiancé Rose which not only accepts me working until 4am but also helps in some steps of home fabrication.

by Marcelo Johann

As you may notice, I am writing the least I can, so this page started like a placeholder. The few sentences below include some details that you are probably not going to find elsewhere. Let´s tell the main part of the story by pictures...
I included a better description of the project at the very end. Check it out!


Early stages:



DIR 9001 does work with TDA1543A in mode 01: 24-bit right aligned.


First sounds comming out


Now CS8414 with Mode 11: local clock drives everything but there is no reclocking. In this mode, CS8414 sends a variable number of bit clocks per sample to keep synchronized to both SPDIF´s rate and local clock. You see a clock divider at right, hex inverter for input and clock buffering, and a shift register at the left to delay the data line by 1 bit transforming left-aligned mode to I2S. This works only with TDA1543 (not Japanese A) I used the exact 64*fs local clock, there is no reclocking, no synchronization problems, but it does not mean this structure is free from clocking problems, as the drift between source and local clocks will cause time distortions, perhaps less frequent.


First SMD prototype being tested repeatedly due to a reset issue


From the lab to the living room status! After many tweaks, I think this summarizes how it sounds: hard to believe!


Here you can see the 66MHz reclocking add-on board, disconnected.
I did not like how it sounded in this design. I will retry it later on with better components.



Here goes a general introduction:

To listen to digital sound, a circuit must first convert a sequence of numbers into an electronic signal, which will be amplified and converted to air waves by a transducer. The SPDIF is a standard employed by the majority of the audio equipment for the transmission of sound data, and encodes both data and clock, being asynchronous. Although there are ways of synchronizing a system by a second cable that carries only clock, most consumer equipment has no such option, and even in a professional studio there are situations with loops where there is no way to select a single piece that has a high performance clock close to it and slaves all other devices. So, most digital audio circuits have to recover the reference clock from an external source, and this is exactly the case for home digital audio reproduction. In this scenario, we want to implement a separate Digital to Analog Converter that synchronizes and listens to an SPDIF connection, so that in the end we can generate a sound signal with much better quality compared to what our digital source
, like a regular US$50 CD our DVD player, can deliver.

Circuit concepts and tricks:

DIR9001 is a modern SPDIF receiver by Texas Instruments that outputs data with a clock jitter rated as low as 40ps RMS for 44.1KHz signals with 512*fs main clock in its PLL. Clock jitter is a very important subject for digital audio reproduction, as it distorts the reference time domain, and its effects are very audible. To complicate it further, the jitter may have different spectrum components, which will translate into different performances on each part of the audio spectrum. TDA1543 and TDA1543A are "economy versions" of a DAC by Philips that are known to sound incredibly well compared to even the best recent designs. The idea of Non Over-Sampling (NOS) comes from the fact that to theoretically satisfy the error requirements in time*amplitude an oversampled converter would have to present clock jitter unatainable in actual scenarios, as first stated by Mr. Kusunoki. To improve jitter, Kusunoki introduced asynchrounous reclocking. In this scheme, a clock many times higher than the bit clock (say, 50 or 60MHz) is used to re-clock the bit clock, slicing it. This new clock absorbs all jitter within its period size, at the same time introducing frequent shifts when the jitter or clock drift is larger than it. Therefore, the result depends on the quality of the incoming clock, the quality of the new high frequency clock, and also on the interactions between both. The idea of NOS is also frequently associated to the concept of filterless DAC. A filter is said to be required after DA convertion to smooth out the stepped waves created by the digital numbers. However, all filters degrade sound quality, as they implement memory processes, with an innertial behavior depending on the previuos samples. This smears out the information in the time domain and decreases time integrity and correlation. On the other hand, if we simply use the DAC with no output filter, the equipment that follows it, most probably the speakers, and finally, the human auditory system, will naturally filter out components above the audio spectrum. Oversampling is used with digital filters and higher bit counts to just implement the filtering, and that is why people mean filterless when writing just NOS. After the DAC chip, a resistor is employed to convert the TDA1543´s current output into a voltage, and this is supposed to beat any other method, as it has the smallest number of components. In its article, Mr. Kusunoki also introduced the idea of using paralleled DACs to statistically reduce erros, among other reasons, what is easily done with the output currents of the TDA1543. Since those ideas first appeared, there has been an increasing number of people implementing their own DACs both as hobby or as small commertial projects. They sound incredibly well, clearly beating up commertial products that are well known and used worldwide by audiophiles and studios. In my design, I wanted also to get rid of the output coupling capacitor, and for that I shifted the whole power supply down as to bias the external ground reference to 3.85V of the original ground, which is the output DC offset recommended by Doede Douma to minimize THD when DACs operate at around 8V supply.

Resulting Sound:

As in most of the other reports, my implementation exhibits the following characteristics:
- an extended sound stage, letting me perceive much better the 3D placement of acoustical recordings;
- reveals low level information that makes the sound become more present, with a better sense of reality;
- better timbre: instruments sound more close to reality and different from each other, more like in an analog recording;
- better bass and punch. It is surprising how much improvement we can get in low frequencies. Part comes from PSU and DC connection, naturally, but other subtle circuit tricks including clocking also change LF a lot, as already reported by many.

Directions:

As I said earlier, I could not get an improved sound by using reclocking, but it was an initial, quick implementation with average components. It seems I got a boost in medium frequencies that made the sound jump outside, but it definitely lost some of the improvements in subtle distiction of instrument sounds and the clarity and extension of bass. I also could not try eight TDA1543s in parallel with confidence. The circuit is sitting open on the floor, may be the noise was already at its limits, so when I hooked up 8 chips, the DIR9001´s PLL very hardly stayed in synch. Therefore, my next step is to redesign the whole power supply. Afterall, Kusunoki´s published schematics had already 90000uF in its PSU, with separate transformers, LRCRC filter, and so... There is a thread in www.diyaudio.com entitled "Building the ultimate NOS DAC using TDA1541A" with over a hundred pages full of discussions, new ideas and points we must consider for improvements, many of them specifically for the TDA1543.

How I am listening:

I am using four Event 20/20 passive monitors together with a pair of Jordan JX92S fullrange and a pair of NorthCreekMusic D2506S tweeters. Everything is currently being driven by a brazilian Nashville Power 70 amplifier. I have also a Hafler Transnova P3000 (-3dB from 0.15Hz to 300kHz), but in the living room I prefer this Nashville for it is smoother, with a response within -3dB from 5 to 150KHz, THD < 0.03%. I know that the bass reflex studio monitors would not be able to provide an accurate judgement for more subtle improvements or decisions, mainly regarding bass, neither will allow my limited 4.5m room, but that is what I got so far. When I talk about improvements, I am comparing the outcome to a Metric Halo ULN-2 firewire Interface which is a studio equipment, and is in the same league as an Apogee or RME converter, if not better, which are all in the range of US$1k to US$2k. So, although I mentioned US$50 players at the begining, this is not the kind of equipment I am comparing against. In fact, I cannot even recall when was the last time I listened to the terrible sound that comes out from these cheap consumer players. It is also worth to highlight that the devices to which I and others are comparing are 20bit, 24bit, 4-16 times oversampling, and it is interesting we can make a 16-bit NOS DAC that sounds better. So, caution to not be fooled here again by the number's game. We still listen to 44.1kHz, 16bit CDs, and even at the studios it is difficult to make a better sound out from equipments that exhibit higher numbers.

Marcelo Johann, 2009
johann at inf ufrgs br