Using a FPGA will allow use of DACs designed to be controlled by a microcontroller. It is easy for a FPGA to talk I2C to setup the DAC chip and to generate any input formats the DAC chips want.

The FPGA can also do a bit of DSP for oversampling. The advantage is that any filter can be used, and their sonic benefits evaluated.

It is difficult to get an idea from reviews on the Net, because people don't review DAC chips, but whole DACs. Comparing two different implementations with different analog output stages etc. can lead to confusion. Besides, the majority of DIY and commercial DAC designs do nothing against jitter ; so they do not evaluate the intrinsic quality of a chip, but instead its jitter sensitivity. For instance, substituing a less jitter-sensitive chip might sound better, even if the new chip is intrinsically worse.

I only cite the non-24 bit chips for the sake of completeness ; the chip I will use must be capable of 24-192 operation.

SACD is useless, so I do not require DSD capability.

See selection table for most characteristics.

Shortlist :
— PCM1704 (TI/BB)
— PCM1794 (TI/BB)
— AD1955 (Analog Devices)
— AKM 4394 (Asahi Kasei Microsystems)
— CS4398 (Cirrus)
— WM8740 (Wolfson)

---

DAC chips cited for the sake of completess and nostalgy :

TDA1543

Concept : a really oldskool passive current divider, only one power supply.
This one is pretty old. The Philips datasheet describes it as "economy version".

TDA1545

Concept : continuous calibration charge-pumps for the 5 MSB ; active current divider for the rest (implemented with 2048 transistors !)
This is the DAC used in my current DAC. It sounds good, but it is only 16 bits, and there are other, better sounding DACs.

TDA1541A

Another cult classic.
Concept : 3x 2-bit dynamic element matching active current dividers for the 6 MSB followed by 10-bit oldskool passive current divider (emitter scaling).

PCM63

Concept : two matched 19-bit DACs. One handles the positive parts of the waveform ; the other handles the negative parts. This minimizes zero-crossing distortion. These DACs are fed from an oldskool R2R current divider.

AD1865

Used in some rather good-sounding (but badly designed) Audio Note gear.
The 4 MSB are segmented (they use 16 independent transistors) while the rest of the bits use a standard R2R ladder.

---

DAC chips considered for inclusion in this project are :

PCM1704

The last of the dinosaurs : multibit design with expensive laser-trimmed resistors ; only one channel per chip ; perfect characteristics, but too expensive. Burr Brown has been meaning to pull it from production for a long time. The best grade (K) is always out of stock...

The internal design is similar to PCM63, with more bits and a higher sampling rate.

"Product is Not Recommended for New Design (NRND)"
"Replaced By: PCM1796"

Portfolio :
— Team Guido Tent's new CD player
— Shanling CD T-100
— Ayre D-1x DVD-Video/CD player (flagship model)
— MOON EVOLUTION SERIES Andromeda Reference CD Player (flagship model). The website shows a shot of the innards of the player, which shows "suboptimal" layout (ie. supply regulators in the clock path, split ground plane etc)
— TEAC Esoteric UX-3 and UX-1 (flagship). ($8500, yikes !)
— Resolution Audio Flagship
— And various other flagship models.

PCM1794

The new TI/Burr Brown flagship model. They call this "advanced segment".
It is an hybrid between Multibit (6 MSBs) and Delta-sigma (the rest of the bits).
It oversamples the signal to 8fs, and the delta-sigma modulator runs at 64fs.

The built-in 8x oversampling filter can be switched between sharp and slow rolloff. It can also be bypassed, in which case the input data shall be sent to the DAC at 4x or 8x rate.
It has no dedicated microcontroller interface, but should be easy enough to control via a FPGA.
It includes a digital volume control (bleh).

Allowable system clock frequencies :
For 48k based sample rates : 24.576 36.864 49.152 73.728 MHz
For 44.1k : 11.2896 16.9344 22.5792 33.8688 MHz
(in bold is my preferred setting, which allows operation at 176.4 kHz (44.1kHz*4).
Mr. Tent supplies 24.576 MHz and 33.8688 MHz but not 22.5792 MHz

— Atoll CD200 (their flagship model, good reputation)
— Many DIY projects on the Net

Farnell stocks it.

AD1955

The flagship model from Analog Devices.
The oversampling can be 2x, 4x or 8x according to the input frequency. This means you don't choose the oversampling factor : the DAC always uses the same internal sample rate (2x 192kHz or 2x 176.4kHz) and oversamples the input sample rate to the internal rate. It can also use an external oversampler, which must respect the same timings.

There is no slow roll-off oversampling option ; the only option is the traditional brick-wall.
It also has some useless features like : DSD, digital volume control.

The datasheet is quite terse. Not much information on the inner workings of the chip here.

AD1853

The AD1853 was the first audio DAC to support the 192 kHz Sample Rate.
It is similar to AD1955, but simpler, and with lesser characteristics.

Interestingly, there is also the possibility of using the current reference pin (Iref) as an analog volume control, with up to 50dB attenuation. Wether or not using this harms the sonics, I have no idea.

Portfolio :

— Audio Aero Capitole 24/192

AD1852

This is basically an AD1852 with integrated I/V conversion and differential voltage output. The datasheet does not specify if the output filter is continuous-time or uses switched capacitors. When in doubt...

AKM 4394

All DAC models from AKM include switched capacitor filters and are therefore disqualified. However, it might be interesting to try them for a listen, since the output stage is quite simple (a wire, actually, or at worst a differential to single ended converter), and the digital interface is just software inside the FPGA, a listening test might be arranged for just the cost of one stereo DAC chip, which is not that expensive.

Portfolio :
— The APL NWO2.5 which is supposed to be one of the best players available. It uses 20 DACs per channel, though.

Wolfson WM8740

Wolfson, of iPod fame.

Oversampling : selectable brickwall or slow roll-off oversampling filter, or bypass oversampling and use an external digital filter.
The sigma-delta modulator works at 128Fs for Fs < 100 kHz ; 64Fs above.
The integrated I/V and analogue lowpass filter has a -3dB point at 192 kHz. It is not specified wether it is switched capacitor or not.

Profusion stocks it, but they don't want to ship less than 5 at a time.

Portfolio :
— Arcam CD33 (a total of four 2-channel Wolfson 8740 DACs ; the marketing blurb sounds like it has been written by an engineer, quite interesting, leaks good facts about their design, these guys seem to know what they're doing) ; also CD93
— Cambridge Azur 640C (dual differential) and Azur 540C (single DAC)
— Some Onkyo, Kenwood top of the line CD players
— The new Rega Apollo (the marketing blurb is hilarious)

---

Rejects

PCM1792

It appears to be the same DAC engine as the PCM1794, but it has a microcontroller-friendly interface and DSD input. However, when deciding not to use the internal oversampling filter, the input signal pinouts change. Awkward.

Due to the presence of all this extra circuitry, this one is possibly a bit noisier than the PCM1794.

However, the operating rate of the delta-sigma modulator can be selected between 64 and 128 Fs ; this should give a boost in sonic quality.
If the oversampling rate of the delta-sigma modulator is selected as 128 fS, the system clock frequency is over 256 fS.

Use of this function enables the designer to stabilize the conditions at the post low-pass filter for different sampling rates. As an application example, programming to set 128 times in 44.1-kHz operation, 64 times in 96-kHz operation, and 32 times in 192-kHz operation allows the use of only a single type (cutoff frequency) of post low-pass filter. The 128 fS oversampling rate is not available at sampling rates above 100 kHz. If the 128-fS oversampling rate is selected, a system clock of more than 256 fS is required.

In other words, 128Fs delta-sigma'ing is usable only for Fs <= 96 kHz. 192 kHz or 176.4 kHz (44.1kHz*4) reverts to 64Fs. Therefore, oversampling in the PC to 192 kHz or 176.4 kHz would negate this advantage this DAC has over the PCM1794 and allow use of the PCM1794's simpler interface and internal circuitry, which should sound better.

Allowable system clock frequencies : same as PCM1794.

Portfolio :
— Audio Aero Prestige CD/SACD player, and Capitole CD player.

PCM1796/8

These appear to be lower grades of the 1792/1794. Not interesting.

CS4398

This is the self-proclaimed flagship from Cirrus.
All DAC models from Cirrus include switched capacitor filters and are therefore disqualified.
A listening test might be interesting. Maybe.

Portfolio :
— TEAC Esoteric series ; these guys are into some pretty crazy shit, like : monaural DACs (yes, you have to buy two), a reclocker based on an atomic clock (yikes), and even sacd transports ! (some of the gear also uses AD1955). However I wouldn't trust guys who fill most of the marketing blurb talking about the weight of the case and the looks of the connectors.
— Sound Blaster Audigy 4 Pro ; no comment.
— Zhaolu D2C
— Aeron DA-10
— Headroom micro-DAC
— Marantz DV9500