totaldac

Principles

Discrete components DAC:
After spending a lot of time on DAC chips, pcm1704 being one of the best I started listening to a discrete R2R DAC made of simple 1% resistors. Musicality was much better than with pcm1704 but some accuracy was still missing. This was the starting point of my ultimate DAC. I spent years and revisions of PCB to improve it and finally use 0.01% VAR Bulk Metal« Foil resistors Vishay Foil Resistors and improvements in the FPGA.

resistances Vishay VAR

Clock and anti-jitter FIFO:
Jitter is very important for sound quality. It is related to the digital source and to the system clock. These days digital sources are sometimes computers so I wanted to make a DAC able to reduce the jitter of the source. An external clock is a common solution but it requires a digital source equipped with clock input, it exists but it is rare and specific. I wanted to make a DAC which can work with any digital sources and any operating systems and softwares when a computer is used. The only solution I could find is a solution rarely used, it is used only in some high end equipment based on FPGA and it uses a buffer memory (FIFO) to store about 10ms of audio data at the digital source rhythm and output a stream at a local oscillator rhythm. A SPDIF receiver such as CS8412 or CS8416 already uses a voltage controlled oscillator but its command changes very quickly to track the input rhythm and so copies partially the jitter of the source. They have to track the digital source rapidly because they have a 80 nanosecond input to output delay whereas my DAC has a 10 mili second input to output delay.

FIR compensation filter:
Non-oversampling DACs are known for their musicality but they all have a problem, the frequency response is not flat and the treble loss is more than 3 dB at 20KHz. This is called sinus(x)/x loss. All DACs with oversampling compensate the sin(x)/x loss in their digital filter. On the TOTALDAC board I didn't want to use oversampling because I prefer non-oversampling DAC sound, but I used a FIR filter to compensate the sin(x)/x loss. It is a short FIR for high frequencies only, so response before impact is short and is not a problem.

Frequency response at 48KHz with and without FIR filter, the response is flat to 20KHz when the FIR is used:

reponse avec et sans filtre FIR

820Hz square wave sampled at 44.1KHz without treble compensation FIR filter:

square wave without FIR

There are no oscillations, it is the pure NOS DAC mode. The horizontal level confirms the 0Hz bass bandwidth.

820Hz square wave sampled at 44.1KHz with treble compensation FIR filter:

square wave with FIR

When the FIR filter is activated some oscillations are visible but the amplitude and the number of oscillations is much smaller than those on conventional DACs. This is part of what explains the natural sound of the DAC.

Power supply:
For an optimum signal to noise ratio the d1 power supply is in a separate box. All voltage regulators for the audio parts are made of discrete circuits (no 3 pin regulator) to minimise the power supply noise floor.

Volume control:
I have made many different volume controls, including high end pots, relays, Shallco + stepper motor, LDR...
The best sound was with the digital volume control made inside the FPGA of the DAC d1, made with a 69bit resolution.
It was the only way not to "listen" to the volume control components.
This doesn't mean that an active preamp can't improve again the sound, for example some customers use the DAC volume and still use their Shindo preamp with a fixed volume.

Don't improve only the source!

Many high end speaker systems use a bass driver included in the speaker or sometimes as an external subwoofer.

The common situation is this one. One amplifier and one speaker cable drive the speaker system.
The filter embedded in the bass speaker will select the bass frequencies only, using usually at least a coil which has a drawback of limiting the damping, can also be microphonic and have other imperfections, so this filter is not totally transparent.
The filter embedded in the mid/high speaker will select the mid/high frequencies only, using usually at least a capacitor which is well known to have sonic limitations, this is why some capacitors are extremely expensive in audio, better than standard capacitor but still not completely transparent.

Five steps are possible to improve a multiway system:

1-use bi-wiring

A cable is used for the mid/high speaker and another cable is used for the bass speaker, from a single power amplifier. The cable impedance and other cable imperfections will not any more be common to bass and mid/high, resulting in better music resolution.

2-use bi-amping

An amplifier is used for the bass speaker and another amplifier is used for the mid/high. The current required for bass will not anymore interact in the mid/high amplifier through its output impedance and its non-linearities. It is also possible to choose an amplifier good for bass and another amplifier good for mid/high. The unused frequencies which are removed by the speaker filters are a waste of power for the amplifier, about half of the total power capacity is wasted in such a system.

3-use an analog active crossover (=active filter)

The passive filters in the speakers are removed. The amplifiers are directly connected to the speaker drivers. Also it allows to remove all suspicious electronics from the active or passive bass cabinet and you can use a good hifi amplifier instead of the integrated amplifier.
Now the amplifier can have a maximum control of the speaker and no power is wasted.
Drawback: The analog active crossover using components like capacitors, op amps and so on can't be totally transparent.

4-use a digital active filter

This time the crossover in calculated digitally instead of using components. It will also be possible to use a delay line on the bass channel or on the mid/high channel to optimise the speaker impulse response, something impossible to do in active or passive analog filters.
Unfortunately most digital filters are more from computer technology with good man-machine interface software but limited jitter and analog performance, so good in theory but not so good for very high end audio, often not as good as the simple passive filter in high end speakers.

5-use a digital active filter integrated in the Totadaldac

This time you get all above mentionned improvements but with absolutely no extra components, no exta conversions, no extra equipments. Only extra calculation is done compared to a standard Totaldac d1-dual, and this extra calculation is done under 69bit inside the DAC processor, just like the rest of the Totaldac processing. As a consequence many limitations are removed and no new problems are added, and you get the new possibility to use delay line to optimise the impulse response as even the speaker manufacturer couldn't do and lastly you can adjust your speaker in your room, on your electronic equipment and maybe according to your tastes and from the listening position using the remote control! This approach combined with one of the best DACs is probably unique!

The Totaldac D2-dual is a two way digital active crossover, it separates bass and mid/high. It can be assembled with the dual DAC option for the mid high channel.

Several d1-dual with the dual DAC option can also be associated to make a two, three or four way active crossover, so extreme performance.

Active crossovers
The DAC uses the R2R principle. It uses discrete resistors like MSBtech and Lavry. It uses a discrete conversion like DCS , Weiss, Emmlabs... Others use the R2R principle but on a chip like Zanden , Audio Note or other DACs using TDA1541 , PCM1702 ou PCM1704 .

Compared to an active crossover BSS FDS-366 or FDS-388 or Behringer DCX2496, the project TOTALDAC has been created specifically for no-compromise hifi systems and the multi-channel volume control has been integrated in the system. The TOTALDAC board is optimised to be able to deal with a digital source (CD, computer...) with only one digital to analog conversion, it is difficult enough to make a state of the art DAC, and this without synchronous or asynchronous sampling frequency conversion. Instead of converting any source to 96KHz or 192KHz 24bit the TOTALDAC board recalculates the delay length and the filter coefficients to adapt automatically to the source.

Measurements
I use several types of measurement equipments but for noise measurement I have built mine which lets me get a better noise floor than I had with a R&S UPL audio analyser. I learned a lot designing it so I have applied this experience to optimize the d1-dual noise floor.

The noise floor is still good enough to measure the d1-dual with enough margin (0dB at 3.3Vrms). The -162dBr discrete signal level is as low as 0.025ÁVrms.

TOTALDAC noise analyser



office telephone: +33 2 90 02 11 39 (european time zone)
mobile telephone: +33 6 18 03 14 08 (european time zone)