Improve Holo Spring3 R2R DAC with FIFO re-clocked DDC streamer transport and SC-Pure clocks

I’m interested in the sound of a NOS R-2R DAC, so I bought a Holo Spring 3. After three weeks break in, I have to say I like the sound more and more.
The output voltages of Spring 3 are RMS 2.9V/RCA and 5.8V/XLR, so the sound levels are a bit higher than standard DAC output levels which are RMS 2V/4V.

Spring 3 is a typical NOS R-2R but with higher accuracy than many others because of the compensation technology.
The sound is very impressive but the style is different from the popular Delta-Sigma DAC, with great transparency, higher density, more details and dynamic, more natural. The 3D image is very solid. The sound is also very clean because of the black background. Those are typical R-2R DAC sound signatures. Spring 3 could be the best R-2R DAC at this price range.

However, when comparing with my well running-in FifoPiQ7/9038Q2M Dual Mono/OPA861I/V DAC system (With 90/98MHz OCXO), What I’m not happy with is that the 3D instrument image of the Spring 3 was pretty big so the whole soundstage was not that pin-point focused and delicate. I will also feel a bit tired after listening to it for a longtime.

Spring 3 uses an internal PLL as the audio MCLK, I believe that could be the reason. Fortunately, Spring 3 has an I2S/DSD over HDMI input option, so I’m wondering if I still have a chance to improve the sound quality of my Spring 3 by feeding better external I2S/DSD signals with higher grade clock to it.

I’m gonna try my new flagship FifoPiQ7 + HdmiPiPro + MonitorPi + PurePi ultra-low jitter network streamer transport solution with my Spring 3.

Somebody said that the R-2R DACs are immune to clock jitter, but it could be not true.

After break-in the Spring3 for more than a week, I started trying to improve it with my external HDMI streamer transport (kind of DDC). Here are the details.

Network streamer transport (DDC) setups:
(#1D) FifoPiQ7 with default 45.1584/49.1520 MHz XOs
(#17B) HdmiPiPro with 8K HDMI cable
(#49A) MonitorPi (optional)
(#48A) PurePi
(#51A) Acrylic protection cover (optional)
RaspberryPi 4B

Spring 3 R-2R DAC configuration:
PLL: OFF
HDMI: ALT2

I user XLR as output, I found set OUT POL: REVERSED spring3 sounds better in this case.

Listening test configurations:
B&W 802 D3 speakers
Pass XA 160.5 pure class A mono block amplifiers
Pass XP-10 Pre-Amplifier in balanced inputs/outputs
PS Audio P20 PowerPlant

The first impression was pretty positive. The R-2R sound signature was still there. But the sound stage opened more than before. The instrument size becomes smaller and also focuses better. The micro details were also improved. However, the 3D image was still not as good as my Q7(with OCXO)/ES9038 Dual Mono/OPA861 DAC setup. So next, I’m gonna try to improve the power supplies and the XO clocks to see how much satisfaction I can get the most out of the Spring3.

Again, the listening test is just a kind of a subjective test. It is based on my own personal experiences. So it only stands my point of view. It would be reasonable if different people feel differently. If it is possible, try it by yourself and trust your ears.

If the clock quality is the most significant thing for a DDC streamer transports, then the power supply would be the most significant thing for the clocks.

1. Add a UcConditioner Pi 3.3 to the HdmiPiPro/FifoPiQ7/PurePi configuration
The result is pretty positive. I clearly noticed more improvement made over the configuration without a UcCoditioner.

2. Replace the UcPure 3.3V battery power supply with a UcPure MKIII 3000F 3.3V ultracapacitor power supply.
UcPure 3000F ultracapacitor power supply could be so far the best ultra-low noise power supply in the real world. I have never found any power supply better than the UcPure.
When upgrading with UcPure, it again makes me realize how important the power supply is. The sound stage is fully open, deep and wide. The instruments sound images are now clearly from 3D pin points. The Spring3 R-2R has neve sounded as good as this.
I was trying to power the HdmiPiPro and the FifoPiQ7 clean side with two independent UcPures, but it didn’t make a real audible difference. It seems sharing the 3.3V power supply would be still good enough if the power supply reaches the level of a UcPure.

Let me explain why the passive UcPure ultracapacitor power supply is better than any active regulators or LDOs.
First, think of what is the working principle of a voltage regulator?
Any regulators are based on the control loops. Basically they are error amplifiers. The control loop uses the difference (error) between the output and voltage reference as a negative feedback to control the output current or the shunt element trying to reduce the error. However, in the real world, both input (AC) and load are changing dynamically. So the output voltage difference will always be there. A good low noise regulator can make the difference smaller but will never eliminate it.

Second, think of where the noises of a regulator come from?
They are the input has noise, the active components have noises, the internal voltage reference has noise, the control loop has noise. And there is no way to eliminate any of them.

Back to the UcPure ultracapacitor power supply. When it works in pure mode, there is no any power input, no active component involved. No control loop and feedback. Only the 230+230uOhm internal ESR. And the 3000F ultracapacitor itself is quieter (like a black hole) than any voltage references in active regulators. I think those are the reasons.

Tipes:
The yellow Eaton 3000F ultracapacitors are so far the best I have ever used
HdmiPiPro really makes difference over the TransportPi and the HdmiPi

Next, I’m gonna try different clocks for Spring 3.

1.To compare phase noise, oscillators have to be at the same frequency.

2.45.1584MHz is selected for the comparison because 45/49 MHz clocks can cover the full digital audio ranges from 44.1 to 384KHz

3.SC-Pure phase noise was measured recently by a calibrated E5052 phase noise analyzer from a third party. But all other phase noise numbers are collected from published datasheets or websites. So please correct me if there is anything wrong.

4.No doubt, phase noise numbers have a big impact on the DAC sound quality, but it may not be 100% related. It could be because of the limitation of the fundamental frequency only phase noise measurement method. DAC runs by the square wave clock, not the sine wave. However, each harmonic frequency has its own phase noise and they all contribute to the final phase noise of the square wave clock. That’s why clocks with the similar phase noise numbers could still sound differently.

5. I could be the one who did the most listening tests to different clock oscillators besides Doede. Based on my own experiences of the listening tests, I've found that the close-in phase nose, the phase noise floor and the phase noise in between (Leeson’s equation applied) are all significant to the sound quality. An oscillator can not achieve a higher grade sound quality if it doesn’t have good close-in phase noise even though its noise floor is great. On the other hand, a good close-in phase noise oscillator can also sound a bit fat, less sound stage any dynamic if the noise floor or the phase noise in between are not great.

6.45.1584MHz Neutron Star2 is a pretty expensive clock. But I found it just uses a fundamental non SC-CUT cut crystal (not the 3OT SC-CUT) . It does have a heater on the PCB but just to keep the crystal temperature stable at around 40 degree C regardless of the ambient temperature. Don’t know why NewClassD didn’t publish the phase noise specifications.

7. Normally SC-Cut crystal oscillators have both better overall phase noise performance and better sound quality than AT-Cut crystal oscillators.

8. Limited by the noise floor of a phase noise analyzer, oscillators with great far off phase noise may not correctly show from the measurement results.

Now it’s time for the real listening test using different clocks. To address the environment EMI noise issue, I set up a Raspberry Pi free streamer/DDC for the test.

RPi free DDC streamer transport (DDC) configuration

#1D FifoPiQ7 with different clocks
#17B HdmiPiPro
#35C StationPi SMT
#49B MonitorPi Pro
#19D ReceiverPi DDC
USB interface

Power supply configuration
#41C UcPure 3.3V at SYNC mode
#41C UcPure 5V at SYNC mode
Spring 3 R-2R DAC configuration
NOS
PLL: OFF
HDMI format: ALT2

Listening test audio configurations
B&W 802 D2 speakers
Pass XA 160.5 pure class A mono block
Pass XP-10 Pre-Amplifier XLR input/output
PS Audio P20 PowerPlant

Music files
Use many different music files. But recommend the original soundtrack of Pirates of the Caribbean (The Curse of the Black Pearl) to feel the good clocks.

Listening testing method:
To eliminate psychological effect, when switch from clock A to clock B, repeat the test using the following sequence
A-B-A-B

Clock comparison testing result