W A R N I N G !

W A R N I N G !

This page is full of non-facts and bullsh!t, (just like the internet and especially forums and other blogs), please do not believe entirely without exercising your intellect. Any resemblance to real things in reality is purely coincidental. You are free to interpret/misinterpret the content however you like, most likely for entertainment, but in no case is the text written on this blog the absolute truth. The blog owner and Blogger are not responsible for any misunderstanding of ASCII characters as facts. *cough* As I was saying, you are free to interpret however you like. *cough*

Thursday, March 8, 2012

Cheap and good or just cheap? LJM / zoe_tsang CS4398 eBay DAC

Update: Read Take Two of this DAC here.

CS4398... the best DAC offering from Cirrus Logic. Sure other companies have even better stuff, like AD1955, PCM1792A, and Sabre, but CS4398 is a good DAC. Xonar DX uses CS4398, and you know how good it is. (Not as good as STX and its PCM1792A duh) Chip-specs-wise, it does -120dB dynamic range and -107dB (0.00045%) TDH+N, which is pretty good already and you rarely see even expensive gear hit that good. (Real equipment worth real money designed for real performance by real engineers often hit that low or better, that's why they use even better DAC chips.)

Ok, so chip specs are rarely actually hit in reality when you put the chip into a badly-design piece of junk. And many products made by obscure companies fall under the "hey lets swap the op-amp for something with 0.00003% THD and replace the DAC chip with better one with -132dB SNR and don't result in any difference because the entire product does worse than -110dB SNR 0.002% THD anyway, but I'm still going to hear the difference because of placebo and I don't ABX for statistically significant results".

(Achieved without even turning on my amp. Hope you are as open-minded towards statistical variance as you are towards the effects of audio "tweaks".)

So, when a designer of a cheap DAC publishes measured results (or so I think) saying 0.001% THD, I get excited. To put things into perspective, EMU0404 (both card and USB) are rated around there (and measures around there), Cambridge DacMagic Plus is rated <0.001% (and should probably hit that if not better), Audioengine D1 is rated <0.002%, NuForce uDAC-2 is rated _____ THD+N (see it on their website), Fiio E17 is rated <0.007% for DAC, ODAC work-in-progress does around 0.003% right now apparently, and the many others without manufacturer-rated specs and the many many with manufacturer-rated specs but does not hit even close when measured which I will not post the links to their measurements or people will again think I have evil agendas. Well, I can't say no, because I am spoiling their sales. In the same way the government is evil because he banned the sales of melamine-tainted milk so the producers are going to starve.

Ok, so 0.001% is good, good for an external DAC (ASUS is spoiling the sound card market), very good for an external DAC board kit that costs $30 USD. But are those measurements real, or did they (or he) measure something else then post it as this DAC's? Lets find out. By buying this thing. And by building this thing.

The construction

After exactly one month after paying the money, the DAC kit made its way from Hong Kong to my residence in Singapore as this pile of parts -

I ordered a green PCB with blue capacitors, and received a blue PCB with green capacitors. Well, China showing you one thing and giving you another is a common thing.

Good quality PCB. I love the way some of the traces are curved, but I feel that some of the traces can be wider. Then again I didn't learn PCB design, so I won't know if it is intentional.

One 3.3V regulator for each digital IC, each regulator getting its 9V from the LM7809. Only one dual op-amp powered by the standard LM7812 + LM7912 combo, and that op-amp is NE5532. One CM102S USB receiver, one CS8416 SPDIF receiver, and one CS4398 DAC. Simple and straightforward, no clock crystal (except for the 12MHz one required for USB), no asynchronous sampling rate conversion, no digital filter, just the 2 chips needed to make things work (3 if you count the USB).

A problem with cheap kits is that the chassis + transformer + connectors are not cheap, so they can reduce the value-for-money-ness and make cost-saving parts choices uneconomical. Fortunately I had a faulty DAC lying around for a project like this.

Construction complete

The measurements

Test setup

Windows 7 @ 80%
Musiland Monitor 01 USD @ 24/96
Belden 1505F RCA
LJM CS4398 @ Zhaolu D1.3 transformer + chassis + connectors
HybridMAX Pro RCA
Fiio E11
Xonar DX

0dBFS output voltage

0.0dB into Xonar DX, or about 2.08Vrms.

Distortion at 0dBFS

This is one of those products that do not see their distortion shooting sky-high at max output level. Even better, this is probably one of those rarer products that perform best at 0dBFS. I mean, 0.00033% THD? This is the first time this system has measured this low, and makes me wonder about the accuracy of ARTA's calculation. The spectrum does say peaks of below -115dB. That said, this figure was only achieved for a few seconds, as the amount of 2nd and 3rd harmonics fluctuates with time. A more realistic value would be a still very impressive 0.0005%. THD+N measurements don't mean much for now, because the 50Hz hum is hurting noise performance, but the current measured value is still a decent 0.0016%.

Difficulty in measuring

My setup has a noise at 7kHz and its 3rd harmonic 21kHz that has been hounding the majority of my equipment. It doesn't show up all the time, though it is there for most of the time, and shows up as a whine if my amp is turned to max. It is there with both sound cards and DACs, it is there in the bedroom and the living room, it is there morning and night, and it is there with an AC EMI filter or not. It fluctuates widely from as high as -90dB to as low as less than -120dB, and I don't know what on Gaia that is. The only time it gets eliminated is when an external self-powered DAC with a floating ground is used with an amp with its ground modded to connect to earth, but simply connecting the amp's ground to earth creates other issues in different situations.

Usually the 7kHz whine is not a problem in measurements, as long as it stays 10dB or more below the highest distortion peak. But with this DAC the highest peak is -110dB. A little bit of noise sends the THD numbers flying from 0.0005% to 0.001%. 0.001% (-100dB) is still very good and not always attained, which is why I don't have problem most of the time.

Hence I recommend looking at the graphs instead of just numbers which are not good indicators of the DAC's actual performance. I will try my best to capture without the noise in ARTA, but I cannot do anything about RMAA.


Unloaded measurements are carried out with DAC @ max volume + Fiio E11 adjusted to give 1.94Vrms into Xonar DX. Xonar DX's included for comparison, in case there is bottlenecking by Xonar DX's input.

LJM CS4398
Xonar DX loopback
THD 20Hz0.0020%
THD+N 20Hz0.030%(A)
THD 1kHz0.00048%
THD+N 1kHz0.0014%(A)
THD 10kHz0.00062%
THD+N 10kHz0.0016%(A)
ASUS's specifications of Xonar DX suggests better input performance than output, and it is showing here.

LJM CS4398 eBay DAC's noise performance is hurt by the 50Hz hum, but THD performance alone is slightly* better than Xonar DX. And at this kind of performance level, slightly* better is worth a lot of money.
 *About 2.5dB better.


Unfortunately I failed to recreate the best measurement I managed to get in RMAA due to the random noise, so here is one that was done at 80/100 volume.

Detailed RMAA results
LJM CS4398 @ 80/100

The DAC's good performance also shows in RMAA, with awesome THD and IMD figures. The random 7kHz and 21kHz noise managed to stay low enough to not affect most of the tests too much. SNR of the DAC is likely to be bottlenecked by Xonar DX's own SNR, as long as the 7kHz noise isn't interfering. The IMD performance is relatively good - other products may also have 0.0005% to 0.0010% THD, but the IMD can  be 0.0020~0.0030%.

However, amidst the great measured numbers, one area performed rather badly - crosstalk. At -80dB, it is onboard sound standard, and far from "SNR minus a few dB" which one expects to get. I wonder if this has anything to do with grounding and other aspects of my build, because at one point I managed to get better than -90dB, and with this DAC otherwise meeting the expectations (something you don't usually see), I'm giving it the benefit of the doubt.

The performance as a headphone amp

I'm going to do a slightly unorthodox approach this time. I manually find the volume before clipping, and measure at that volume.

Detailed RMAA results
LJM CS4398 loaded

The op-amp filter cum buffer is pretty nicely built, don't let the size (and component count) fool you. Into 300Ω it delivers excellent distortion values again at near maximum output level. Crosstalk, the only bad number on this DAC, is hit and drops further.

Into 16Ω the DAC pushes an ok 0.4V with an ok performance before clipping, could be better but I've seen amps that do worse, in either output level (into 16Ω) or distortion or both.

Output impedance... a crude measurement limited by the accuracy of my multimeter + the dB reading which only goes down to 0.1 precision. I'm estimating around 1Ω, which is pretty overkill for a DAC out and better than some headphone amps out there.

Note that these results do not mean that you should be using this DAC as a headphone amp; insufficient voltage for high-impedance loads, lack of volume control potentiometer for preserving SNR at low output voltages (needed for high-sensitivity and/or low-impedance loads), and poor crosstalk performance. But when you are in a pinch, this thing can drive headphones decently while within its limitations.

The subjective listening

I would like to point out that the opinions written in this section were formed before measurements were taken, or if the opinions are of the DAC after certain modifications, then they were formed before new measurements for the modified DAC were taken. This is to eliminate placebo that arises when you know the performance of the equipment.

I would also like to point out, time and time again, that these observations have not been double-blind-tested and would most likely fail such testing, so the accuracy of the statements is pretty much nil and you are to believe those words at your own risk.

Actually I don't know why I still write the subjective listening. Time and time again I want to just heck it and don't give people any wrong thing to spread. But I'm doing this because people want them. Because people don't bother trying to understand what the numbers mean. Because they can't save themselves or hundreds of their dollars even if you shove information in their faces. Because people prefer reading vague words instead of dead numbers. I still write because... I just don't know...

Right after construction completion, the sound was pretty nasty and devoid of extreme frequencies. After a night of power-on, the sound settled into something more normal.

The high frequencies has to be the best part of this DAC. Slightly relaxing, but wide and open and a good feeling of separation. There is also more high frequency content, but not the tone control kind of more treble, rather due to hearing each instrument and each sound following through and dying off independently. The statement that digitized music will never feel like live seems like a lie. Not an easy feat with the Far East Asian music I listen to. (It's easy with well/intentionally-mastered tracks, I know there are tracks of water dripping, but do you listen to music or water dripping?) If it was a little bit more aggressive, it would be perfect.

Impressions of mids are mixed. On one hand the background instruments are heard nicely and not getting killed by the prima donna, but on the other hand there is not enough warmth from harmonious chords. Also there seems to be some kind of strange distortion, but that could have been due to the air conditioner. (Fan blowing wind, which when mixed with moving air from the speakers, can cause issues. The easiest way to hear the effect of this is with a sine test tone. With a fan blowing around you, the sine wave turns into a high-speed mini dubstep. Wub wub wub wub wub wub...)

Bass, can't exactly say good or bad. Extreme deep bass seems to be missing, but there is a good amount of the rest. Again not the tone control kind, of bass, but with good separation. It's like, you are the guy playing a bass instrument, you hear the bass louder, you concentrate on the bass more, and the bass notes are evenly reproduced throughout the frequency range, unlike some setups that emphasize on certain frequencies only and you hear certain notes louder than it should while other notes go missing.

On the whole, a nice, spacious, relaxing DAC to listen to, people who want warm and phat look elsewhere.

The cheap

A USD$30 DAC would have it flaws. And no the flaws are not with regards to the sound quality, measured or otherwise.

Flaw #1 - The SPDIF input is not transformer-isolated

A cost-saving measure perhaps, because magnetics are expensive. But not using the isolation transformer brings to this DAC a problem which is present on computer sound cards and computer-powered USB DACs - noisy ground. And to finish the combo this DAC does not have optical input which would've provided the isolation.

"But the SPDIF standard also calls for an isolation transformer on the source side too", I can hear you saying. However the reality is that majority of computer sound cards, especially the motherboard integrated ones, do not use an isolation transformer.

Here is a comparison of measured performance when using onboard Realtek vs Musiland Monitor 01 USD as coaxial source:

With the computer's ground connected to the DAC's ground at the SPDIF connector, the resulting ground loop (the DAC's ground connects back to sound card's ground) hurts the noise level, which in turn brings performance down to onboard sound standards. What might be news to you readers, is that there are many sound cards and DACs with this problem. As just mentioned, sound cards and non-isolated DACs suffer from this issue, whether it shows up or not depends on your entire setup's grounding configuration, and even if it shows up whether you hear it or not depends on where you control the volume. Combined with the inability of sound cards to use potentiometers to reduce volume and keep noise level low (because reducing volume in the digital domain kills signal-to-noise ratio), this results in the semi-true myth that computer sound cards are noisy, in the face of measurements revealing -110dB noise levels. Sound cards have good signal-to-noise ratios, but only under the right conditions, which is when the sound card is being tested.

Solution? Use an SPDIF source with an isolation transformer. Like this -

Or this -

Bottom line, I won't exactly flame LJM since there are other products with the same problem and this problem is situational, but not following standards is not following standards (EDIT: Ok, after much digging, I found that only the EBU (as in the guy in AES/EBU) version of the standard requires mandatory use of transformer for both transmitter and receiver,  while AES3 does not. So mistake on my part.) and this can cause issues with low-budget users who are likely to buy this DAC and are likely to use it computer motherboards' SPDIF outputs, an issue that IMO can reverse the decision to buy this product.

Flaw #2 - Does not handle no-input well

While not a performance issue, this can be devastatingly irritating. While most other products simply keep quiet when the input is switched off or disconnected, this thing goes into a hiss. This can be a major issue for people who keep their DACs on all the time, have sources that switch off SPDIF output automatically (sends nothing, not even 0, which in SPDIF is a square wave), and/or have their amps at full volume and use the player/OS to control volume. Imagine you turn off the source or some funny thing happens, and a loud hissing is blasted from your speakers. Not good.

I am not sure if this is a design fault or cost-cutting, but this is a rather unacceptable flaw, considering a proper product should not have this problem.

The ugly

- The USB input isn't really useful, being limited to 16-bit/48kHz. I'm not going to test that, because the best 16-bit can do is 98dB dynamic range, and this DAC's distortion peaks are lower than that. Not a big issue since competing products at that price range don't have 24/96 USB either. This is one of the cheapest decent DAC kits anyway, so you don't get to choose.

- Requires a transformer, or 2 DC power supplies or a power supply with ±V out, either way it is relatively expensive given the cheap price of the DAC.

- No optical input. Besides providing the much-needed isolation, there are more laptops with optical out and virtually zero with coaxial out. That said, many laptops don't have optical out either. (MacBooks have)

The verdict

At USD$30, it is cheap for a DAC board kit. At US$124 for a finished product, it is still cheap for an external DAC of that performance. However with two significant flaws, it is hard to recommend it to others, especially those that just started out in audio, have a low budget, and have limited accessories and knowledge. But if you can look past its flaws and have the accessories to back it up, you are looking at an awesome product.

No comments: