384Khz ‘NOS’ Dual AD1955. The 3D-Stereo Natural DAC.

I like to make things, and I’ve always been interested in how things work. When I was 10, I had saved enough pocket money to buy a transistor radio. Within a week I had taken the back off to marvel at it. I’m still doing stuff like that today.

If you are new to electronics, the contents of this page will bore you to death. Maybe more than once !?!  To me, it is a record of the great pleasure of taking something very good and making it exceptional.

I got this DAC in November 2012 and I liked the sound very much, so I thought I’d share about how it works and how I’ve made it work better. I finished changing it in May 2014.

After modification

After modification. Click for a closer look.

It’s a dual Analog Devices AD1955, (it has two AD1955 chips, one for the left and one for the right) with a “smooth interpolation” asrc (asynchronous sample rate converter ) to 384khz/24bit.  So whatever the digital source, let’s say CD, it takes the CD sample rate of 44.1Khz 16bit and upsamples it to 384/24 using an algorithm. This means, when it is converted to analogue, it will sound more natural and less digital.

The asrc is an Altera Cyclone III FPGA and the signal from this is fed into the dual AD1955 for independent left/right DAC, and the AD1955’s internal oversampling filter is disabled. This is what makes this DAC very special: there is no oversampling done in the DAC, so it is an upsampling non-oversampling (NOS) DAC.

And it sounds really good !

The two AD1955 have their own independent left/right analogue 5V supplies from two LT1763 regulators. Two LT1461 and two LT1761 provide the independent left/right Vrefs to balance out any DC offset. They are all good regs with low noise etc so I’m very happy with that, especially considering the price, and the design doesn’t use any coupling capacitors which would degrade the sound.

It had six opa627 op amps which sounded very smooth and airy, but I changed the buffers to lme49990ma, and they seem to be almost completely transparent. I’m also using ADA4627br in the I/V, which gives it a more dynamic sound. Further changes led me to use ADA4898 in I/V and AD797 in the buffers.

Other op amps I’ve tried :
FET opa827, opa1641, opa602BP
BJT OPA211D, LME49710HA, LT1028, AD8597, OPA1611,

Left offset is typically around -6.9mV and right is 3.9mV, depending on op amps of course. There’s no trimming in the circuit, just as the AD1955 datasheet suggests.

Less impressive are the 7809/7812/7912 analogue supply regs and long supply lines.

Each 7809 reg supplies 40mA to 3 ICs: LT1763 & LT1461 & LT1761. They share their supply with the op amps.

The 6 single-channel op amps have independent left and right supplies too – but they use 7812/7912 only, supplying +/-14mA with the current op amps.

With the current caps and transformers, before regulation, I get +/-21V with 50mV ripple on +21V and 20mV ripple on -21V. For the digital circuit, I get 8.95V with 210mV ripple.

The AD1955s digital supply is shared from one central ams1117 5v reg.

The caps in the I/V // buffer feedback circuit are marked as Evox Rifa.

It sounded really good as it was but I couldn’t help but try to upgrade it.

First of all, I changed the electrolytic caps. The digital now uses an Elna RJH 4700uF 16V and this has cut ripple to 120mV. The analogue is all Elna SilmicII – 4 x 1000uF 25V and 4 x 470uF 16V and 2 x 47uF 25V. That’s given it a fuller smoother sound but the ripple has increased to 120mV on the +ve rail and 40mV on -ve.

The second change is the rectifier diodes. Out with the 1n4007 and in with MUR810 + 220pf X7R snubbers across the legs. The bass got a lot more articulate and clear, and I must admit I was quite surprised by how much difference this simple change made. Neg rail is -21.5V with 66mV ripple, and pos is 21.1V and 141mV. I wanted to get lower ripple so I also added two more Elna SilmicII 1000uF 35V caps, so finally it is -21.7V/45mV, and +21.3/114mV.

Like every change, I gave these changes several days of constant use before doing anything else, just to be sure the changes were improvements.

Third change was the xo. I lifted the ferrite input to the xo and added a 1uF X7R after it to create an LC filter. I also added a cap multiplier/potential divider to drop the output of the 5V AMS reg to 3.3V (measures 3.33V with CCHD-957 xo). The divider is 1.2k and 5.1K, and the cap multiplier is Oscon 100uF and BC550C. Sound quality was fantastic but there was better to come.

Fourth change is the clock circuit’s power. I replaced the 1n4007 rectifier diodes with BYW80 and 220pF snubbers. I replaced the AMS1117-5V reg with ADP7104. It’s output noise is 15uV compared to 150uV for the AMS. Neither datasheet advises about output impedance but the adp seems to have reasonable transient response. Ripple rejection is similar, maybe slightly better for the adp7104.

To reduce ripple and isolate the clock power from the rest, the input to the adp7104 is via a 100R resistor and a Nichicon 470uF PLF low esr cap. This RC filter drops 0.5V (measured; so there’s a 5mA load it seems; cchd-957 datasheet says 15mA ?) and it reduces ripple by about 100mV (about 30dB). Measured input ripple on the input to the adp7104 is 3mV, so the output of the adp7104 should be at its noise floor of 15uV, which means the clock should be at the noise floor of the BC550, less some dB at high frequencies for the LC filter.

The adp7104 also feeds the AMS1117-3.3V for the 74VHC14, and this was replaced with an ADP151, 9uV noise, 3.3V regulator. To ensure stability, I added 1uF X7R across C905 and C909, and a 0.1uF across C907.

The result was far more detail and depth to the sound. Actually, it was quite amazing. I gave the clock circuit a week to burn in before making more changes.

The next change was replacing the 7805 that is a pre-regulator for Altera asrc board. I replaced this with a BUP41 150Mhz pass transistor with a filtered ref02 5v reference at it’s base (filtered through an rc ; 1K + 47uF).

This gives 4.4V under load (measured at 89mA at the input to the AZ1084S + load through the ams1117 2.5V reg) into the Altera board, which is just enough for the dropout on the board’s regulators (1V). There is no measurable ripple after this pre-regulator. I also replaced the 220uF Nippon cap with a 820uF Nichicon PLG low esr cap.

The BUP41 is dropping 5.6V @150mA (estimated) so it has a heatsink. I measured tempuratures over 2 hours and it sits at 44 deg C.

Change number 6 – the ams1117 3.3v reg that powers the HCU04 and the optical receiver. I lifted the leg and measured current drain – 23mA. I removed the ams and added an RC filter 56R + 820uF PLG cap , followed by a 78L05 and 10uF tantulam on the 78L05 output, and then an adp151 3.3V reg, with a 1uF X7R cap after the adp151.

Change 7 – the ams1117 5V for the AD1955 dvdd supply was replaced with another RC filter (15R and 820uF) followed by a 78L06, then and adp7104 5V reg. The rc drops 0.56V so the pair of ad1955 are drawing 37mA; datasheet says 20mA each. It also reduces 110mV ripple to 14mV, which is 17.8dB, exactly as most calculators say it should, and helps to isolate the supply from the other components. An example calculator :


I chose a smaller value (15R) here because I wanted to make sure the 78L06 pre-regulator had over 2V drop out. When I fitted it, I tried to keep the reg, cap etc away or above the data lines as much as would allow.

That last change made a very noticeable difference – more precise and natural sound.

Next,  I removed / moved some of the tantalum caps and fitted 1uF x7R where necessary to follow the advice in the adp151 datasheet; match input and output capacitance and minimum 1uF ceramic x7r.

The 7809 have been replaced with a 78L15 & ADP7104 9V. I didn’t hear any change in sound quality from this, perhaps just a little more bass. The 78L15 has only 30dB ripple rejection but combined with the adp, the output should be at the adp’s noise floor of 15uV. The 78L15 is dissipating 7×0.04=280mW and rises to a case temp of 52 degrees in a 22 degree ambient. It’s too hot to hold a finger on, but well within its soa for long-term use, as is the adp.

The op amps all have yellow tantalum capacitors on their supply rails. The markings suggest they are AVX 22uF 16V. There are twelve of these around the op amps, decoupling their supplies. I have replaced them with SilmicII 10uF 16V.

Purely for cosmetic reasons, I changed the LEDs. They are common cathode so I used a bi-colour red/white for power/lock LED and red/green/blue tri-colour for the input selection LED.

I replaced the clock distribution IC – a Fairchild 74VHC14 – with a Potato 74G14. It sounded worse at first so I removed the RC filter on the output ( no C, shorted R) and this did not restore the previous sound – it improved it. However, I had expected better than this from my previous experience so I then tried a clock fanout buffer, the ON NB3L553, which has just 29fs of additional phase noise. This worked very well and gave the vocals a smoother edge with more clarity, space and definition overall.

Finally, I fitted the XMOS USB to I2S board, with an isolator and a clean power supply for the DAC side of the isolator. After that, I replaced the op amp regulators with Ti TPS7A4901 regulators for lower noise. This improved the stereo image and detail. I wish I had done this earlier – those 78XX and 79XX op amps regs are far inferior. So, with that done, I dod some more op amp rolling and settles on OPA627 in the I/V and AD797B with 47pF bypasses in the buffer. So out came the op amp sockets and the ICs were soldered in.

And, it’s finished ! It’s really hard to fault this DAC in any way now. It’s the best sounding DAC I’ve ever heard. Eat your heart out ES9018, you’re number 2. Yes, the ES9018 is more punchy but this has the most overall pleasing sound I’ve heard.

Amazing Sound Starts at the DAC

Amazing Sound Starts at the DAC