• Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

On the Distortion of Cirrus Logic CS431xx-Based Devices: A Comparative Review

How would this review influence your purchase decision of a device employing Cirrus Logic CS431xx?

  • Going forward I will not buy a device if it adopts any Cirrus Logic DAC chip.

    Votes: 7 8.5%
  • I would not consider any device with CS431xx.

    Votes: 10 12.2%
  • I'd consider a device with CS431xx only if it's been tested free of the "Cirrus hump" distortion.

    Votes: 42 51.2%
  • I don't care about this distortion issue and would just consider the device's other features.

    Votes: 23 28.0%

  • Total voters
    82

jkim

Senior Member
Forum Donor
Joined
May 23, 2024
Messages
386
Likes
803
Location
US
TL;DR / Changelog
  • [5/20/25] Eleven DAC/HP amp combo devices adopting either the Cirrus Logic CS43131 or CS43198 were tested, and seven of them were found produce the distortion dubbed a "Cirrus hump." Based on measurements from multiple sources (including this review), a list of Cirrus-Logic-based devices was compiled.​
  • [5/26/25] Part II results of another testing method revealed that even the devices tested free of the Cirrus hump distortion produce distortions of a similar nature in a certain, restricted condition (when the signal level is increasing).​
  • [5/27/25] A list of recommended portable DAC/HP amp devices, alternative to CS431xx-based devices, was added to Additional Remarks after Part II Tests.​
  • [5/31/25] Standard performance measurements of the reviewed devices were added to Appendix B.​
  • [6/3/25] Although not part of this review, it was found that the headphone output of Apple MacBook products, adopting a custom Cirrus Logic DAC chip, also suffers from Cirrus hump distortions, and reported in a separate post.​
  • [7/5/25] Some listening tests and comments on the audibility issue were added in a separate post.​
  • [7/7/25] A workaround for devices supporting the NOS (non-oversampling) mode, which prevents the distortion, has been found and reported in a separate post.​


Introduction

PXL_20250520_201925734.MP~2.jpg

Since their release in fall 2017, Cirrus Logic's two digital-to-analog converter (DAC) chips, the CS43131 and CS43198, have gained traction in industry, being adopted in quite a few consumer hi-fi and professional audio devices. With the recent boom of hi-fi products designed and manufactured by Chinese companies, more and more products adopting these chips keep coming out these days. Attractive features that have led these manufacturers to use the chips are:
  • Excellent, if not state-of-the-art, measured performance in SINAD/THD+N and dynamic range
  • Relatively low cost
  • Integrated, quite capable headphone driver (see Appendix A; even the CS43198 has a headphone driver)
  • Small packaging for wide applications in portable devices
  • Energy-efficient
Despite this popularity, some industry engineers and DIY audio enthusiasts have criticized these DAC chips. One issue is about the technology called Dynamic Range Enhancement (DRE) implemented in the chips. This technique increases a DAC's low-level performance and achieves a higher measured dynamic range (DR) by adjusting the chip's gain structure, depending on the signal strength, to find a good trade-off between digital and analog gain settings. DRE is not a new technology but has also been employed by other DAC chip manufacturers. Setting aside the debatable point on whether the DR of a DRE-enabled device, measured by the AES17 standard procedure, is faithful to a theoretical definition of a dynamic range, criticism against these Cirrus Logic chips has been on the fact that the use of DRE is not officially documented by Cirrus Logic anywhere.

Another issue, relatively less known, is that some devices employing these chips exhibit substantially elevated distortion/noise responding to certain signals. This phenomenon was examined and reported in an article published at the Reference Audio Analyzer, a Russia-based audio gear review website founded by Roman Kuznetsov. What this article demonstrated is that this peculiar behavior is not just observed in artificial testing signals but also in real audio content. And it occurs to several CS431xx-based devices tested at the site, except for only a few others producing no such distortion. Based on these observations and the information on the chips' datasheets---i.e., based on what he can see---, he conjectures that the problem could be caused by a malfunction of the chips' Class H amplifier and would be resolved by proper firmware design.

The purpose of writing this review at ASR is multifold:
  1. To test more CS431xx-based devices toward compiling a list of devices affected by this problematic distortion behavior;
  2. To suggest an effective measurement method for detecting and characterizing the distortion;
  3. To eventually (and hopefully) have audio manufacturers react and come up with a solution (if they are going to keep using these chips).
The purpose of this review is not to find the precise cause of the problem or suggest a solution, simply because it is not possible to achieve based on measurements alone or the known information.

Measurement Setup

The Peculiar Distortion: Cirrus Humps

Some CS43131-based portable devices tested here at ASR showed excellent noise and distortion performance as well as very nice output power. To name a few: MEIZU Hifi Audio Pro, Tanchjim Space, TempoTec Sonata BHD Pro, JCALLY JM20, etc. In fact, many CS431xx-based devices are nearly indistinguishable when tested by a standard package of measurements, unless incorrectly designed (e.g., inadequate caps in power supply circuits). Problems occur when a complex signal (e.g., multitone) is played at somewhat low levels.

Let's take some multitone measurements from my recent review of the FiiO KA15:
index.php


These are the FFTs of a 32-tone signal at the maximum, unclipped level as well as at a 6 dB lower level. Normally, if we feed the same signal at a lower level, the ratio of the signal to distortion/noise deteriorates only slightly because of the weaker signal, just as shown above. However, the FIIO KA15 does not exhibit such an incremental trend but an abruptly high distortion when the signal further weakens:
index.php


And the response becomes normal when the signal reaches a much lower level:
index.php


The entire picture of level-dependent distortion can be seen in the sweeping measurements of Total Distortion + Noise (TD+N) using the same signal over varied strengths:
index.php


Notice a clear "hump" which indicates the rise of TD+N in a range of signal strengths. Let us dub this a "Cirrus hump." In comparison, the JCALLY JM20 and JM20 MAX do not show this behavior in the same measurements.

One may think this measurement signal, made of 32 equal-amplitude tones, is unrealistic which may never occur in real audio content. In fact, essentially the same phenomena can be found with any multitone signals, even with dual tones, as long as tone components are in similar amplitude. See below:
index.php


This dual-tone signal, called "TDFD Bass" in Room EQ Wizard, is simply composed of 41 Hz and 89 Hz sine tones. As with multitone signals, the KA15 distorts responding to these dual tones in a range of signal strengths. Again, a sweep of these tests across a range of signal strengths was performed:
index.php


A hump in this sweep test of TD+N versus output levels is unmistakably clear, too, compared to other CS43131-based devices (JCALLY JM20 and JM20 Max) which do not produce distortion.

Distortion Measurements of CS431xx-based Devices

An ASR member kindly sent quite a few CS431xx-based devices for testing. And I also own a couple. Here is a list of tested devices:
PXL_20250520_201925734.MP~2.jpg

(From left to right)
  • FiiO BTR13 - QCC5125, dual CS43131, no op-amps
  • FiiO KA15 - SPV5048-Pro, dual CS43198, dual SGM8262-2
  • HiBy FC5 (w/ display) - Airoha AB1565, dual CS43198, dual SGM8261-5
  • JCally JM20 - SA9312, single CS43131, no op-amps
  • JCally JM20 Max - SA9312, single CS43131, SGM8262-2
  • JCally JM20 Pro - CB5100-Pro, single CS43131, no op-amps
  • JCally JM28 - SA9312, single CS43198, no op-amps
  • Shanling UA1 Plus - CT7601, dual CS43131, no op-amps
  • Tanchjim Stargate II - CB5100-Pro, single CS43131, no op-amps
  • TRN Black Pearl - CB5100-Pro, dual CS43131, no op-amps
  • TTGK TT39518F01-Pro module - CB5100-Pro, dual CS43198, no op-amps
These were measured exactly in the same condition. The following devices were found exhibit the distortion behavior:
  • FiiO BTR13
  • FiiO KA15
  • HiBy FC5 w/ display
  • JCally JM20 Pro
  • Tanchjim Stargate II
  • TRN Black Pearl
  • TTGK TT39518F01-Pro
And the following did not produce distortion:
  • JCally JM20
  • JCally JM20 Max
  • JCally JM28
  • Shanling UA1 Plus
See their multitone sweep tests:
32-tone_level_sweep_results.png

Note. This distortion behavior does not depend on the device's playback sample rate, impedance load, or any firmware settings like digital filters, Class H/AB modes, etc. (if available)---it always occurs regardless of such conditions. The Fiio BTR13 was not included here due to its 16-bit resolution masking distortion to some extent (instead, see below for its TDFD bass distortion measurements).
Across the affected devices (thick solid lines) distortion profiles over the multitone signal level, i.e., Cirrus humps, are consistent with each other. The TD+N results of unaffected devices, shown in dashed lines, are just as expected, being dominated by each device's power of noise components.

Below are the sweep results of dual-tone TDFD Bass IMD test signals:
TDFD_Bass_Level_Sweep_Results.png

Note. As with multitone tests, the above distortion behavior does not depend on the device's playback sample rate, impedance load, or any firmware settings like digital filters, Class H/AB modes, etc. (if available). The Fiio BTR13's noise level is much higher than other devices' due to its 16-bit resolution.
Again, the elevated TD+N in a range of signal levels is an unmistakable character of the affected devices (thick solid lines), whereas the unaffected devices' measurements mainly indicate their noise components (dashed lines).

Other Characteristics

The devices affected by the above measured distortion respond even to a single tone in a similar manner if the tone frequency is 8 Hz or below. This was first observed by an ASR member and reported here. All the affected devices reported above were found produce the same distortion in response to this subsonic sine tone whereas unaffected devices do not (like this). Another interesting characteristic discriminating the affected devices from unaffected ones is their noise level. For all CS431xx-based devices, their noise level depends on the test signal level in a distinctive manner due to the adopted DRE (dynamic range enhancement) technique.

See below for the effect of DRE in action on the noise level:
RMS_Noise_v_Level_v2.png


What is interesting is that, for all the devices showing the distortion behavior, this noise level is a simple function of the test tone level as shown above. It does not depend on the previous signal levels. This is not the case of the devices not producing distortion. See below:
index.php

Note. The same behavior as above was first observed by @torgeirs from the Fosi Audio DS2 and reported in the DS2 thread (see posts #780 and #783).
That is, the noise level depends on not only what test tone level a sweep test starts from but also in which direction the test goes from there. If the signal weakens incrementally, the noise level stays the same as at the starting point, meaning that the DRE logic does not increase its effect, until the signal finally reaches below -50 dBFS. But if the signal strengthens, the noise level adapts to it meaning that DRE decreases its effect (by trading digital gain for analog gain). This implies that DRE in unaffected devices makes less aggressive adjustments of its gain structure (combination of digital & analog gain). Less active DRE, on whatever logic it operates, is not a problem at all, because the noise performance of CS431xx-based devices would by no means be poor even without DRE.

Of course, based on these observations alone, we cannot draw a conclusion on why the unaffected devices do not produce distortion. But we may conjecture that this different operation of DRE may be related to the occurrence (or absence) of the distortion.

Audibility of the Distortion

It is difficult to tell how clearly audible these measured distortions would be in real audio content. But given the fact that the problem is observed even in a simple dual-tone test reported above, it is no wonder Roman at RAA was able to easily spot a movie soundtrack to demonstrate the distortion. Note that the website provides multiple recordings of the same track played at different levels, which are then level-matched for the higher % distortion to be heard easily. Other than the most audible cases, it takes trained ears familiar with distorting sound to hear it. It is not clear "clicking" or severe clipping distortion. And the original track already contains quite high background noise. The distortion sounds like occasional crunch in the midst of fluctuating noise floors. In comparison, through an unaffected device, the background noise is not much fluctuating without crunch.

But we are talking about 0.06% distortion in the worst case. And human ears are easily deceived in this case. That is one big reason why the issue of subjective audibility is difficult to resolve. Also, there must be infinite possibilities of this distortion occurring in audio content. That is another reason why this issue should not be considered based on subjective listening of a few samples, and why systematic measurements---even a simple sweep of test tones over varied signal levels---are important.

EDIT (7/5/2025). As time has passed and more tests have been done, I feel the need to update my take on the audibility issue of this distortion. To avoid a verbose description of this topic, this new content was put in a separate post.

Conclusion

Tests of eleven DAC/headphone amplifier combo devices that adopt either Cirrus Logic CS43131 or CS43198 were performed, and seven of them were found exhibit the same distinctive distortion behavior as reported in the article published at the RAA website. It was also demonstrated that level sweep tests with multitone signals can be used to detect and characterize this distortion behavior. Also found was the unique operation of DRE (dynamic range enhancement) in unaffected devices different than in affected devices.

Setting aside the issue of subjective audibility, this measured distortion is obviously an engineering flaw that should not occur in DA converter products targeted at Hi-Fi markets. It is really bothersome that any device employing CS43131 or CS43198, whether it is a headphone dongle, desktop DAC/HP amp, digital audio player, or a music streamer, should be suspected to potentially produce this distortion. Based on the fact that some devices are not affected by this problem, like Roman at RAA I also believe it can be resolved by firmware design. But the problem is that there is no clear register in the chips' datasheets that must be related to this behavior. No one knows for certain, perhaps except for Cirrus Logic engineers. For the reasons listed earlier, we will see more and more products with CS431xx will keep coming out. We hope audio gear manufacturers will react to this issue and find a solution, and better yet, it will become common knowledge.

Lastly, based on the measurements by RAA, myself and others, a spreadsheet of Cirrus-Logic-based devices has been compiled to show devices tested producing and not producing distortion, and other information. This list will be updated as more devices are tested.

Additional Remarks after Part II Tests

Running all the tests including those in Part II (see below) has shaped my take on this issue. Of course, my perspective on this hobby has also influenced it. Listeners may not easily notice the distortion examined in this review since it is not severe clipping. However, as indicated by the testing conditions (i.e., simple multitone, even dual-tone, signals in a wide range of low amplitude levels), this distortion occurs not just in very limited, artificial situations. It should occur frequently in real audio content, although it is another matter whether a listener can hear it or not. Human auditory perception is gullible in isolated situations like hearing distortion in a controlled condition, but at the same time it is subtle and sensitive holistically. For this reason, I am still not comfortable.

But perhaps a more important consideration is that hi-fi consumers of these devices would enjoy listening to music through them believing that they are exceptionally transparent by modern standards, when in fact not. That is robbing them of big fun in this hobby!

Another serious problem is that these Cirrus Logic chips are dominating the portable DAC market. So much so that it is difficult to find a portable DAC/HP amp that does not use CS431xx and is at the same time nice on paper. Alternatives are rare. There are some products that adopt ESS ES9039Q2M or ES9069Q (or previous gen ES9038Q2M) or AKM AK4493S. But many of inexpensive options are paired with RT6863/SGM8261 op-amps which are not very good (Shanling UA2 Plus, UA3, UA4, or UA5; or Celest CD-20). If someone concerned with the distortion of CS431xx asks to recommend a reasonably priced portable DAC/HP amp, then the following devices come to mind (for now):
  • Qudelix 5K: Now that the problem of CS431xx is known better, the Qudelix 5K is even more appealing. Sure, it's more expensive but still reasonably priced and supports Bluetooth as well as on-board parametric EQ (PEQ). What more would you ask for?
  • FiiO BTR15: Potentially a good device for the same reason as the Qudelix 5K is. But given the superior user interface and functionality of the 5K, I would not choose the BTR15 over the 5K. Edit. Had a chance to measure the PEQ response of the BTR15 as a USB DAC. Far from being correct. I do not recommend this product if its PEQ function is needed for USB connection.
  • FiiO BTR17: Highest output power for a portable device. Also with on-board PEQ and Bluetooth support. But substantially more expensive than the 5K or BTR15, and somewhat bulky for a portable device.
  • FiiO KA17: Similar to the FiiO BTR17 except for no Bluetooth support. Still a bit pricy.
  • iFi Go Link Max: If you don't need Bluetooth or PEQ, this seems to be a good choice.
  • Creative Sound Blaster X1 (same as the former model SXFI Amp): For the current price ($45-$50 as of June 2025), this USB DAC/HP amp is a worthy contender. Relatively high unbalanced output of 2 Vrms into 32 Ohms and 2.4 Vrms into 300 Ohms. Based on quality components like the AKM AK4377 (DAC chip) and TI INA1620 (op-amp). Supports 10-band graphical EQ and has useful features for microphone users like noise cancellation, auto-mute, mic mute button, mic EQ, etc.
  • Hiby FC3 or FiiO KA1 or iFi Go Link or Hidizs SD2: If you don't need Bluetooth or PEQ or higher-power balanced output, this seems to be a good budget choice.
  • Neutron HiFi DAC V1: A niche product with on-board DSP functions like PEQ, cross-feed, and surround sound. Represents the best implementation of the ES9219 DAC chip. Unbalanced output only. No Bluetooth.
  • E1DA 9038D: A niche product designed by E1DA (Ivan) using ES9038Q2M, providing extremely transparent unbalanced 3.5 mm output. Just a USB DAC. No on-board PEQ, no Bluetooth.
  • E1DA 9039S: Another niche product based on ESS's recent ES9039Q2M, boasting measurably the most transparent output with extremely low noise and distortion even under low impedance loads. Just a USB DAC. No PEQ, no Bluetooth. It provides only a 2.5mm balanced connection which should limit usability for some users.
  • Topping DX1: Not truly portable but small enough to carry in a bag. Based on the excellent AKM AK4493S chip. Just a USB DAC, no on-board PEQ, no Bluetooth.


Part II: Another Testing Method

Another interesting, notable observation has been made since the above review was posted. An ASR member, @danadam, reported that some distortion is still generated from a CS431xx-based device that is tested not producing the "Cirrus hump." To understand how this distortion can occur, note that all the distortion measurements reported in the original review were based on an FFT analysis of a test signal in a steady state (i.e., the signal is played at a fixed level). And according to my conjecture, described in the section "Other Characteristics" earlier, DRE (dynamic range enhancement) is a suspected cause of the distortion and is in action even for the unaffected devices. In particular, its adverse effect may be limited to a situation in which the signal level is monotonically increasing. In this case, distortion may occur because the digital gain needs to be reduced but may not be done in a timely manner as the signal level increases. Inspired by this review, @danadam came up with a test method in which the amplitude of a multitone signal is changed over time. Using this signal, the JCally JM20's response was recorded and a spectrogram analysis (FFTs over time) of the recording was performed. Indeed, distortion was observed in a brief time period in which the signal level was increasing (see posts #37 and #40). In addition, the measurements of the Tanchjim Space (post #42) suggested that a device producing a Cirrus hump would exhibit distortion over the entire playback, i.e., regardless of the signal level increasing or not.

To confirm these findings, some devices included in the review were tested again using the proposed method. The test signal consists of a chord with 32.7, 40.9, and 49.1 Hz components (CMaj in just intonation) with its amplitude changing over time as shown below in waveforms:
index.php


This track was played from a DUT (device under test) with its Fs set to 96 kHz. The playback level was -15 dBFS at which the Cirrus hump distortion should be most noticeable. And response was recorded by the Cosmos ADCiso with its Fs set to 192 kHz. Lastly, the recording was level-matched for its peak to be -1.0 dBFS. The resulting spectrograms of tested devices are shown below.

JCally JM20 / JM28 (tested not producing a Cirrus hump):
JM20_CMaj.png


The FFT analysis here used a Hann window with a 2k length for distortion to be easily spotted. Note that distortion, indicated by long thin vertical lines, occurs when the signal level is increasing initially but once it becomes steady the distortion no longer occurs.

Essentially the same results were obtained for other devices with no Cirrus hump. For example,
JCally JM20 Max (tested not producing a Cirrus hump):
JM20Max_CMaj.png


Now, how about the devices tested producing Cirrus humps? Here are some examples.
Tanchjim Stargate II / JCally JM20Pro (tested producing Cirrus humps):
JM20Pro_CMaj.png


It produces distortion for each and every signal spike, whether the signal level is increasing or not, when the level is in the affected range.

TRN Black Pearl (tested producing a Cirrus hump):
TRN_Black_Pearl_CMaj.png


Basically the same result for the TRN Black Pearl.

How about a device not using a CS431xx? The Qudelix 5K with an ESS DAC chip (ES9218P/ES9219C) was taken out.
Qudelix 5K:
Qudelix_5K_CMaj.png


Clean. No hint of distortion.

The original test track and recordings of the tested devices' playback are attached in two ZIP files (File 1 and File 2) if you want to listen for yourself. To my ear, distortion from the devices with no Cirrus hump (JCally JM20/JM28 and JM20 Max) is hardly noticeable. But distortion from the ones with Cirrus humps (JCally JM20Pro/Tanchjim Stargate II and TRN Black Pearl) sounds like subtle crunch along with fluctuating background noise.

So, what can we say? It is still difficult to tell about the audibility of distortion from CS431xx-based devices in real audio content. In particular, distortion from devices not producing a Cirrus hump would be unnoticeable in most situations, given that it is low-level and should be infrequent.

What would be a solution? If possible, I suggest DRE should be disabled entirely. The noise performance of CS431xx-based devices would not be poor even without DRE. Of course, an ultimate solution should be a better DRE algorithm, but this cannot be done without developing a new chip. By the way, we know DRE is not documented in Cirrus Logic datasheets. However, we see one suspicious register appear on the CS43131 datasheet:

CS43131_DS1155F2_pdf.png


And this parameter (DRE_EN) is NOT described anywhere in the datasheet. We want it to be disabled! But wait. The dynamic range with DRE_EN = 0 is more than 20 dB lower than with DRE_EN = 1? If this is true, what a bummer!
EDIT. I missed one thing in the above datasheet. The listed dynamic range for either DRE_EN = 1 or 0 is so low simply because they used the measurement bandwidth 20 Hz - 90 kHz which included the noise shaping effect above 50 kHz. The DAC's dynamic range in 20 Hz to 20 kHz is decent even without DRE.


Appendix A: CS43131 versus CS43198

The topic here is not about the two Cirrus Logic DAC chips' distortion but about their difference, if any. The block diagrams on Cirrus Logic's corresponding product pages indicate that the CS43131 integrates a stereo headphone amplifier whereas the CS43198 does not. The current drive capability of many CS43131-based devices with no additional op-amps has been shown to be very nice. How about a device employing only one CS43198 or two with no headphone driver added? We may think it will serve only as a line-level DAC source. However, if you search AliExpress, you will see some headphone dongles described as having a CS43198 or two with no op-amps. A headphone dongle with no amplifier? It actually makes sense ... because the CS43198 shows no difference from the CS43131 in measurements! Let's consider two pairs of devices that are believed to be clones of each other:
  • The JCALLY JM28 is believed to be a clone of the JCALLY JM20 except that the former is with a CS43198 and the latter with a CS43131.
  • The TRN Black Pearl is believed to be a clone of the TTGK TT39518F01-Pro except that the former is with dual CS43131's in a nice black case and the latter is with dual CS43198's in a bare module.
PXL_20250520_202120197.MP~2.jpg
PXL_20250520_202209583.MP~2.jpg
PXL_20250520_203336682.MP~2.jpg
PXL_20250520_203041942~2cropped.jpg


It was confirmed that the two DAC chips on the TT39518F01-Pro module have letter markings "CS43198."

The measurements of THD+N versus output voltage under 300 Ohm and 32 Ohm loads for these clone pairs were found identical to each other:
CS43198_THDN_v_Level_300Ohm.png

CS43198_THDN_v_Level_32Ohm.png

Note. In the case of TTGK TT39518F01-Pro and TRN Black Pearl, their unbalanced 3.5mm outputs were measured.

If headphone drive capability is identical between the two chips, then what tells them apart? I cannot think of a reason other than a marketing scheme.


Appendix B: Evaluation of the Devices Against Standard Performance Metrics

Since this review tested many similar devices, it would be remiss if their performance were not evaluated with respect to conventionally reported measurements. This information would be especially valuable to those who believe the Cirrus hump distortion should be mostly inaudible, on which I tend to agree.

A succinct and effective way to see the performance of a DAC/HP amp is to perform sweep tests of two kinds: THD+N vs. output level (using 1 kHz sine tones) and THD+N vs. frequency. In the results shown below, only unbalanced 3.5mm outputs were measured for all devices even if some provide balanced outputs. This was to make the measurements comparable across them because unbalanced outputs usually have lower noise levels than balanced counterparts at the same output level.

THDN_v_Level_32Ohm.png

THDN_v_Frq_20kBW_300Ohm.png


Key takeaways:
  • Most of these CS431xx-based devices' THD+N measurements are similar to each other, representing excellent performance.
  • Although devices with additional op-amps can provide higher output levels, CS431xx's max output even without an op-amp is still very nice.
  • Some implementations without op-amps show slight signs of stress under the low impedance (32 Ohm) load, shown in their THD+N vs. output voltage: JCally JM20/JM28 (worse) > Tanchjim Stargate II/JCally JM20Pro > TRN Black Pearl/TTGK TT39518F01-Pro (no increased THD). Still no audible concerns, though.
  • The Hiby FC5 produces elevated THD in response to higher-frequency tones (thin green line on the THD+N vs. frequency plot), which is most likely due to the characteristic of its adopted op-amp chip (SGM8261-5).
  • The unique design of the Shanling UA1 Plus employing dual CS43131s for its single-ended 3.5mm output does not seem to improve noise or distortion performance (thin Cyan lines on both plots). It may have somewhat improved output power into lower impedance (< 32 Ohms), though.
  • I would choose the JCally JM20 Max as the overall best performer for unbalanced output. It maintains low noise & distortion while achieving higher power with the help of an op-amp (SGM8262-2).
  • The TRN Black Pearl/TTGK TT39518F01-Pro with no additional op-amps is very nice, too. To feed dual DAC chips, this larger device must have incorporated better power supply circuits than dongle-type devices.
 

Attachments

Last edited:
Hey @jkim, thanks for posting. :)

However if you wouldn't mind editing your post in line with the below please;


JSmith
 
Hey @jkim, thanks for posting. :)

However if you wouldn't mind editing your post in line with the below please;

JSmith
Thanks for pointing that out. Edited. And of course, I did not mean it to be derogatory at all, as you may have guessed already.
 
Introduction

View attachment 452324

Since their release in fall 2017, Cirrus Logic's two digital-to-analog converter (DAC) chips, the CS43131 and CS43198 have gained traction in industry, being adopted in quite a few consumer hi-fi and professional audio devices. With the recent boom of hi-fi products designed and manufactured by Chinese companies, more and more products adopting these chips keep coming out these days. Attractive features that have led these manufacturers to the chips are:
  • Excellent, if not state-of-the-art, measured performance in SINAD/THD+N and dynamic range
  • Relatively low cost
  • Small packaging for wide applications in portable devices
  • Integrated, quite capable headphone driver (see below; even the CS43198 has a headphone driver)
  • Energy-efficient
Despite (or because of) this popularity, some industry designers and DIY audio enthusiasts have criticized these DAC chips. One issue is about the technology called Dynamic Range Enhancement (DRE) built into the chips. This technique increases a DAC's low-level performance as well as achieves a higher measured dynamic range (DR) by adjusting the chip's gain structure, depending on the signal strength, to find a good trade-off between digital and analog gain settings. DRE is not a new technique but has also been employed by other DAC chip manufacturers. Setting aside the debatable point on whether the DR of a DRE-enabled device, measured by the AES17 standard procedure, is faithful to some theoretical definition of a dynamic range, criticism against these Cirrus Logic chips has been on the fact that the use of DRE is not officially documented by Cirrus Logic anywhere.

Another issue, relatively less known, is that some devices employing these chips exhibit substantially elevated distortion/noise responding to certain signals. This phenomenon was examined and reported in an article published at the Reference Audio Analyzer, a Russia-based audio gear review website founded by Roman Kuznetsov. What this article demonstrated is that this peculiar behavior is not just observed in artificial testing signals but also in real audio content. And it occurs to several CS431xx-based devices tested at the site, except for only a few others producing no such distortion. Based on these observations and the information on the chips' datasheets---i.e., based on what he can see---, he conjectures that the problem could be caused by a malfunction of the chips' Class H amplifier and would be resolved by proper firmware design.

The purpose of writing this review at ASR is multifold:
  1. To test more CS431xx-based devices toward compiling a list of devices affected by this problematic distortion behavior;
  2. To suggest an effective measurement method for detecting and characterizing the distortion;
  3. To eventually (and hopefully) have audio manufacturers react and come up with a solution (if they are going to keep using these chips).
The purpose of this review is not to find the precise cause of the problem or suggest a solution, simply because it is not possible to achieve that based on measurements alone or the known information.

Measurement Setup

The Peculiar Distortion: Cirrus Humps

Some CS43131-based portable devices tested here at ASR showed excellent noise and distortion performance as well as very nice output power. To name a few: MEIZU Hifi Audio Pro, Tanchjim Space, TempoTec Sonata BHD Pro, JCALLY JM20, etc. In fact, many CS431xx-based devices are nearly indistinguishable when tested by a standard package of measurements, unless incorrectly designed (e.g., inadequate caps in power supply circuits). Problems occur when a complex signal (e.g., multitone) meets with somewhat low levels.

Let's take some multitone measurements from my recent review of the FiiO KA15:
index.php


These are the FFTs of a 32-tone signal at the maximum, unclipped level as well as at a 6 dB lower level. Normally, if we feed the same signal at a lower level, the ratio of the signal to distortion/noise deteriorates only slightly because of the weaker signal, just as shown above. However, the FIIO KA15 does not exhibit such an incremental trend but an abruptly high distortion when the signal further weakens:
index.php


And the response becomes normal when the signal reaches a much lower level:
index.php


The entire picture of level-dependent distortion can be seen in the sweeping measurements of Total Distortion + Noise (TD+N) using the same signal over varied strengths:
index.php


Notice a clear "hump" which indicates the rise of TD+N in a range of signal strengths. Let us dub this a "Cirrus hump." In comparison, the JCALLY JM20 and JM20 MAX do not show this behavior in the same measurements.

One may think this measurement signal, made of 32 equal-amplitude tones, is unrealistic which may never occur in real audio content. In fact, essentially the same phenomena can be found with any multitone signals, even with dual tones, as long as tone components are in similar amplitude. See below:
index.php


This dual-tone signal, called "TDFD Bass" in Room EQ Wizard, is simply composed of 41 Hz and 89 Hz sine tones. As with multitone signals, the KA15 distorts responding to these dual tones in a range of signal strengths. Again, a sweep of these tests across a range of signal strengths was performed:
index.php


A hump in this sweep test of TD+N versus output levels is unmistakably clear, too, compared to other CS43131-based devices (JCALLY JM20 and JM20 Max) which do not produce distortion.

Distortion Measurements of CS431xx-based Devices

An ASR member kindly sent quite a few CS431xx-based devices for testing. And I also own a couple. Here is a list of tested devices:
View attachment 452324
(From left to right)
  • FiiO BTR13 - QCC5125, dual CS43131, no op-amps
  • FiiO KA15 - SPV5048-Pro, dual CS43198, dual SGM8262-2
  • HiBy FC5 (w/ display) - Airoha AB1565, dual CS43198, dual SGM8261-5
  • JCally JM20 - SA9312, single CS43131, no op-amps
  • JCally JM20 Max - SA9312, single CS43131, SGM8262-2
  • JCally JM20 Pro - CB5100-Pro, single CS43131, no op-amps
  • JCally JM28 - SA9312, single CS43198, no op-amps
  • Shanling UA1 Plus - CT7601, dual CS43131, no op-amps
  • Tanchjim Stargate II - CB5100-Pro, single CS43131, no op-amps
  • TRN Black Pearl - CB5100-Pro, dual CS43131, no op-amps
  • TTGK TT39518F01-Pro module - CB5100-Pro, dual CS43198, no op-amps
These were measured exactly in the same condition. The following devices were found exhibit the distortion behavior:
  • FiiO BTR13
  • FiiO KA15
  • HiBy FC5 w/ display
  • JCally JM20 Pro
  • Tanchjim Stargate II
  • TRN Black Pearl
  • TTGK TT39518F01-Pro
And the following did not produce distortion:
  • JCally JM20
  • JCally JM20 Max
  • JCally JM28
  • Shanling UA1 Plus
See their multitone sweep tests:
View attachment 452227

Across the affected devices (thick solid lines) distortion profiles over the multitone signal level, i.e., Cirrus humps, are consistent with each other. The TD+N results of unaffected devices, shown in dashed lines, are just as expected, being dominated by each device's power of noise components.

Below are the sweep results of dual-tone TDFD Bass IMD test signals:
View attachment 452228

Again, the elevated TD+N in a range of signal levels is an unmistakable character of the affected devices (thick solid lines), whereas the unaffected devices' measurements mainly indicate their noise components (dashed lines).

Other Characteristics

The devices affected by the above measured distortion respond even to a single tone in a similar manner if the tone frequency is 8 Hz or below. This was first observed by an ASR member and reported here. All the affected devices reported above were found produce the same distortion in response to this subsonic sine tone (like this) whereas the unaffected devices do not (like this). Another interesting characteristic discriminating the affected devices from unaffected ones is their noise level. For all CS431xx-based devices, their noise level depends on the test signal level in a distinctive manner due to the adopted DRE (dynamic range enhancement) technique.

See below for the effect of DRE in action on the noise level:
View attachment 452223

What is interesting is that, for all the devices showing the distortion behavior, this noise level is a simple function of the test tone level as shown above. It does not depend on the previous signal levels. This is not the case of the devices not producing distortion. See below:
index.php


That is, the noise level depends on not only what test tone level a sweep test starts from but also in which direction the test goes from there. If the signal weakens incrementally, the noise level stays the same as at the starting point, meaning that the DRE logic does not increase its effect, until the signal finally reaches below -50 dBFS. But if the signal strengthens, the noise level adapts to it meaning that DRE decreases its effect (by trading digital gain for analog gain). This implies that DRE in unaffected devices makes less aggressive adjustments of its gain structure (combination of digital & analog gain). Less active DRE, on whatever logic it operates, is not a problem at all, because the noise performance of CS431xx-based devices would by no means be poor even without DRE.

Of course, based on these observations alone, we cannot draw a conclusion on why the unaffected devices do not produce distortion. But we may conjecture that this different operation of DRE may be related to the occurrence (or absence) of the distortion.

Audibility of the Distortion

It is difficult to tell how clearly audible these measured distortions would be in real audio content. But given the fact that the problem is observed even in a simple dual-tone test reported above, it is no wonder Roman at RAA was able to easily spot a movie soundtrack to demonstrate the distortion. Note that the website provides multiple recordings of the same track played at different levels, which are level-matched for the higher % distortion to be heard easily. Other than the most audible cases, it takes trained ears familiar with distorting sound to hear it. It is not clear "clicking" or severe clipping distortion. And the original track already contains quite high background noise. The distortion sounds like occasional crunch in the midst of fluctuating noise floors. In comparison, through an unaffected device, the background noise is not much fluctuating without crunch.

But we are talking about 0.06% distortion in the worst case. And human ears are easily deceived in this case. That is one big reason why the issue of subjective audibility is difficult to resolve. Also, there must be infinite possibilities of this distortion occurring in audio content. That is another reason why this issue should not be considered based on subjective listening of a few samples, and why systematic measurements---even a simple sweep of test tones over varied signal levels---are important.

CS43131 versus CS43198

Now the topic is not about these two Cirrus Logic DAC chips' distortion but about their difference, if any. The block diagrams on Cirrus Logic's corresponding product pages indicate that the CS43131 integrates a stereo headphone amplifier whereas the CS43198 does not. The current drive capability of many CS43131-based devices with no additional op-amps has been shown to be very nice. How about a device employing only one CS43198 or two with no headphone driver added? We may think it will serve only as a line-level DAC source. However, if you search AliExpress, you will see some headphone dongles described as having a CS43198 or two with no op-amps. A headphone dongle with no amplifier? It actually makes sense ... because the CS43198 shows no difference from the CS43131 in measurements! Let's consider two pairs of devices that are believed to be clones of each other:
  • The JCALLY JM28 is believed to be a clone of the JCALLY JM20 except that the former is with a CS43198 and the latter with a CS43131.
  • The TTGK TT39518F01-Pro is believed to be a clone of the TRN Black Pearl except that the former is with dual CS43198's in a bare module and the latter is with dual CS43131's and housed in a nice black case.
View attachment 452319View attachment 452320View attachment 452321View attachment 452326

It was confirmed that the two DAC chips on the TT39518F01-Pro module have letter markings "CS43198."

The measurements of THD+N versus output voltage under 300 Ohm and 32 Ohm loads for these clone pairs were found identical to each other:
View attachment 452318
View attachment 452317


If headphone drive capability is identical between the two chips, then what tells them apart? I cannot think of a reason other than a marketing scheme.

Conclusion

Tests of eleven DAC/headphone amplifier combo devices based on either of Cirrus Logic CS43131 and CS43198 were performed, and seven of them were found exhibit the same distinctive distortion behavior as reported in an article published at the RAA website. It was also demonstrated that level sweep tests with multitone signals can be used to detect and characterize this distortion behavior. Also found was the unique operation of DRE (dynamic range enhancement) in unaffected devices different than in affected devices. Lastly, the CS43198, known to be with no integrated headphone amplifier, was found not just capable of driving headphone loads but also produce the same measured performance as the CS43131.

Setting aside the issue of subjective audibility, this measured distortion is obviously an engineering flaw that should not occur in DA converter products targeted at Hi-Fi markets and consumers. It is really bothersome that any device employing CS43131 or CS43198, whether it is a headphone dongle, desktop DAC/HP amp, digital audio player, or a music streamer, should be suspected to potentially produce this distortion. Based on the fact that some devices are not affected by this problem, like Roman at RAA I also believe it can be resolved by firmware design. But the problem is that there is no clear register in the chips' datasheets that must be related to this behavior. No one knows for certain, perhaps except for Cirrus Logic engineers---or even worse, former Wolfson engineers. For the reasons listed earlier, we will see more and more products with CS431xx will keep coming out. We hope audio gear manufacturers will react to this issue and find a solution, and better yet, it will become common knowledge.

Lastly, based on the measurements by RAA, myself and others, a spreadsheet of CS431xx-based devices has been compiled to show devices known to and not to produce distortion, and other information. This list will be updated as more devices are tested.
Fascinating! Thanks for the hard work!
 
The Fiio BTR13's noise level is much higher than other devices' due to its 16-bit resolution.
Why would 16-bit resolution cause higher noise level? Or is it due to the dithering of a 16 bit input signal?

Side note: I thought the BTR13 USB input—like all these QCCxx chips—was UAC 1.0 compliant: it should be capable of 24 bit / 96 kHz…
 
Why would 16-bit resolution cause higher noise level? Or is it due to the dithering of a 16 bit input signal?
Of course, bit depth affects the noise level: 16 bit = SNR 98.09 dB. I actually added 32-bit dither to all test signals. In the case of the BTR13, adding 16-bit dither makes the noise floor a bit higher.

Side note: I thought the BTR13 USB input—like all these QCCxx chips—was UAC 1.0 compliant: it should be capable of 24 bit / 96 kHz…
Unfortunately Fiio limited it to 16 bit / 96 kHz and hence its usability as a USB DAC.
 
Last edited:
Truly a fantastic research you made, thank you for the hard work!
Nothing much to add, if anything I'd like to make a call to everyone reading if they have any information to contribute about the USB bridge of the devices listed AS AFFECTED in the spreadsheet , just to clear up some conjectures @jkim and me made about a possible relation between the "hump" issue and the implemented usb bridge.
Awesome!
 
Thanks, this is good and valuable work. Would it be possible to also test the ramp-up behavior from the standby mode? It's been observed by some folks that some of these DACs/USB bridges like to cut off the first second of audio out of silence, and it's hard to tell which of these dongles are good in this regard.
 
Thanks, this is good and valuable work. Would it be possible to also test the ramp-up behavior from the standby mode? It's been observed by some folks that some of these DACs/USB bridges like to cut off the first second of audio out of silence, and it's hard to tell which of these dongles are good in this regard.
I can do that if you can suggest sort of standard test condition and method. foobar2000 is my go-to music player on my PC. Would you?
 
I can do that if you can suggest sort of standard test condition and method. foobar2000 is my go-to music player on my PC. Would you?
It could be tricky to measure that behavior.
I made some practical tests some times ago with Sonata BHD that has the ramp-up issue:


so, even after finding a valid test measurement, if for example the measuring software behaves like ocenaudio keeping the playback device always active, you wouldn't be able to replicate the issue during measurements.
 
Great testing, I had just got the JCally JM20 Pro for its hardware PEQ and assumed its implementation would be in line with the already tested JM20 and Max. oh well. not sure how to explain JCally putting out designs in the same product family both with and without the CS distortion, perhaps it is something unobvious like which USB controller bridge the design employs. and with the amount of commercial off the shelf designs and integrations available, perhaps JCally doesn't really know either.
 
It's been observed by some folks that some of these DACs/USB bridges like to cut off the first second of audio out of silence, and it's hard to tell which of these dongles are good in this regard.
I can do that if you can suggest sort of standard test condition and method. foobar2000 is my go-to music player on my PC. Would you?
As for “the symptom that the first second of sound is cut off from the silent state,” I think it is easy to identify it because the cut off sound is output by simply pressing the test button in the sound settings of Windows and playing the test sound.

I was using JM30 (CS43131+SA9123L), which has the same specifications as JM20 except that the LED color changes depending on the format (PCM or DSD) being played, on my PC (Windows 11), but I gave up using it because the first part of the sound was cut off without mercy, including the test sound above.
 
As for “the symptom that the first second of sound is cut off from the silent state,” I think it is easy to identify it because the cut off sound is output by simply pressing the test button in the sound settings of Windows and playing the test sound.

I was using JM30 (CS43131+SA9123L), which has the same specifications as JM20 except that the LED color changes depending on the format (PCM or DSD) being played, on my PC (Windows 11), but I gave up using it because the first part of the sound was cut off without mercy, including the test sound above.
I do understand this is a serious issue. When I get a chance, I will do some tests although the condition is limited to one particular setup (Windows 11).
 
Why not saying who so we can thank him too? My bet is on @mc.god...
Cut offs like this are usually the audio device being asleep due to reacting to power saving on the link and the software stack playing audio before the device has woken up and/or losing buffered audio as it has failed to hand it off to the device.

‘Retail’ Windows 11 normally only does USB power saving like this when on battery but OEMs can and do create their own default power plans which switch it on when on AC power.

On the USB front MS USB mice used to be terrible for not surviving power saving, not USB but Intel Macs aggressively put their audio devices to sleep causing pops and cut offs on the output for example QLab grew a ‘keep core audio awake’ option due to my reports about this.
 
Cut offs like this are usually the audio device being asleep due to reacting to power saving on the link and the software stack playing audio before the device has woken up and/or losing buffered audio as it has failed to hand it off to the device.

‘Retail’ Windows 11 normally only does USB power saving like this when on battery but OEMs can and do create their own default power plans which switch it on when on AC power.

On the USB front MS USB mice used to be terrible for not surviving power saving, not USB but Intel Macs aggressively put their audio devices to sleep causing pops and cut offs on the output for example QLab grew a ‘keep core audio awake’ option due to my reports about this.
Your quote of my text makes no sense to me and seems fully unrelated.
 
Your quote of my text makes no sense to me and seems fully unrelated.
May refer to these “cutoffs”:
 
I am probably building a DIY cs43131 headphone amp for one of my projects, if I find schematics to add an output op-amp, then might have the chance to test what settings make what.
Thanks for gathering all this information, will be useful.
 
Back
Top Bottom