What is baudline?
Mystery Signal
Edirol UA-25

vendor Edirol
product UA-25
interface USB 1.1
duplex full
channels 2
resolution 16 bits input and 24 bits output
max rate 48 Ksample/sec
chipset unknown
codec unknown
operating system Fedora core3 Linux x86_64 2.6.12-1
driver snd_usb_audio ALSA 1.0.9rc2
buffer size input 64 KB, output 64 KB
baudline version 1.07
test date Apr 16 2008
notes Limiter off. MON.SW off.

Advance mode off with ALSA at 44100.  Advance mode on with JACK 0.101.1 at 44100 and 48000 for 24-bit Tone Generation.  Not tested at 96000 because the UA-25 is only half duplex at that sample rate.

Used two 0.5m balanced 1/4" loopback cables.

Set the Tone Generator's digital gain to 0 dB.

Vendor=0582 ProdID=0073 Rev= 1.07

This card is part of the Full Duplex DAQ comparison survey.

Sample Rate
The sample rate on DAQ cards is not a fixed absolute constant.  Like time, it fluctuates, and it is difficult to measure accurately.  Sometimes there are relationships between the input and output sample rates that can reveal interesting details about the inner working machinery.

The following table of measurements use a technique described in the sample rate stability application note. The rate column is the sample rate value that the collection hardware is programmed to.  The in/out rate and in/out error columns are absolute measurements of the ADC / DAC clock.  The loop error column uses a tone generator loopback method for a high accuracy measurement of the relative difference between the ADC and the DAC clocks.  The three error PPM columns are theoretically related by the formula: "in_error - out_error = loop_error"

rate in rate out rate in error out error loop error
44100a  44108.671  44108.671  +196.62 PPM  +196.62 PPM  +0.0000 PPM 
44100j  44108.642  44108.642  +195.96 PPM  +195.96 PPM  +0.0000 PPM 
48000j  48009.388  48009.388  +195.58 PPM  +195.58 PPM  +0.0000 PPM 

Rate suffix: a = ALSA and j = JACK

The UA-25 has a loop error measurement of +0.0000 PPM which means that the ADC and DAC clocks are locked.

Frequency Domain
The USB audio device's input and output jacks are connected with short external balance (differential) cables and run in full duplex mode.  This is a loopback test and baudline's tone generator is the signal source.  Distortion, noise floor, filter response, and inter channel crosstalk are the frequency domain measurements of interest in this section. 

The signal test sources are a pure sine wave, a linear sine sweep, and WGN.  The sine wave is used for the distortions and crosstalk measurements.  The linear sine sweep and WGN are used for the filter characterization measurement.  Both are an application of the swept sine vs. WGN technique and are equivalent measures of the frequency response. 

Since spectral performance is a function of sample rate, each of the sound card's native rates will be tested.  The highest sample rate is usually the cleanest and this is advantageous because it allows the isolated testing of the ADC and the DAC.  The matched, source, and sink sample rate combinations are described below.

The input and output sample rates are the same.  This combination tests the performance of both the ADC and the DAC in a matched mode of operation.  The linear sine sweep signal in the left spectrogram display and the WGN (orange) in the Average window characterize the in-band filter response.  The sine wave (green) in the Average window is used for distortion and crosstalk measurements.  The sine leakage (purple) is used for crosstalk measurement

44100a -> 44100a 44100j -> 44100j  
48000j -> 48000j  

Rate suffix: a = ALSA and j = JACK

Clean looking spectrogram sweeps for all the native rates and ALSA/JACK modes.  The peak-to-peak filter ripple seen in the Average 8X window is very low and less than 0.5 dB.  The 11 strongest distortion peaks seen in the Average window are all harmonics of the main fundamental Hz test tone.

The following table of measurements were made using the technique described in the sine distortion application note.  It is a full duplex test that uses a loopback of the tone generator to measure the various distortion parameters.  The stereo crosstalk column is a measure of channel leakage that uses a sine wave channel and a silent channel as the signal sources.

rate SNR THD SINAD ENOB SFDR crosstalk
44100a  +88.77 dB  -86.24 dB  +84.31 dB  +13.711 bits  +90.58 dB  -82.09 dB 
44100j  +88.50 dB  -85.87 dB  +83.98 dB  +13.656 bits  +89.00 dB  -82.15 dB 
48000j  +88.93 dB  -85.31 dB  +83.74 dB  +13.617 bits  +87.89 dB  -82.16 dB 

Rate suffix: a = ALSA and j = JACK

Good distortion measurements with consistent performance at both the 44100 and 48000 native sample rates.  The UA-25 has a strange warm up behavior from the idle (off) state that takes about 5 minutes and increases the ENOB by about 0.6 bits.  Also, the best ENOB is difficult to reproduce since it can vary by almost .1 bits as a function of time.

With the 44100 rate it is interesting that the ALSA performance is slightly better than the JACK performance.  The opposite would be expected since JACK is running the UA-25 with 24 bit samples and ALSA is running with 16 bit samples.  With JACK, baudline 1.07 is inputing 16 bit samples and outputing 24 bit samples with the Tone Generator.  This hybrid 16/24 bit operation should be superior to straight 16 bit.  This performance discrepancy could be due to the UA-25 internal Advance mode operation or it could be due to baudline input dithering.  In any case, for the best performance, it is recommended to run the UA-25 with Advance mode off (16 bit).

This test should be run again in the future when a 24 bit baudline is available to see if the performance of the UA-25's 24 bit Advance mode improves.

periodic glitch
After a warm up period of about 10 minutes it was observed that the UA-25 running at a 44100 sample rate captures a weak wideband spectral glitch.  This glitch is periodic and repeats every 164 seconds after the first occurance.  See the baudline spectro image on the right for an example of what the glitch looks like.  This periodic glitch happens at all sample rates, happens with both ALSA and JACK, and it affects the left and right channels equally.  Using a pure sine wave test signal the glitch is easily visible in the frequency domain.

The UA-25 has a very useful digital loopback button that sends the input channel a pure digital copy of whatever is being output.  To engage this UA-25 feature simply disconnect any optical SPDIF input and press the DIGITAL IN button.  After testing the periodic glitch was still present when the UA-25 digital loopback was enabled.  This signifies that the periodic glitch problem lives in the digital domain which focuses the attention away from the UA-25's analog circuitry. 

Baudline was reconfigured with a 1575 Hz sine wave which is an even multiple of the 44100 sample rate.  This choice of test signal frequency locks the sample values and makes it easier to see any transitory sample damage in the time domain.  Below is the Waveform image that shows the glitch at the 9596 ms position (focus on the positive cycle top where the arrows are pointing).  It looks like a single sample was dropped. 

This sample dropping glitch could be a USB audio driver problem or it could be flaw in the UA-25's data path buffering circuitry.  So the question is whether the Linux USB audio driver or the Edirol UA-25 is at fault?  A future investigation should verify if this periodic glitch exists on a computer with a different USB chip that is using a newer version of the Linux kernel (USB audio driver).

Note that dropping one sample every 164 seconds at a 44100 sample rate is an error of -0.14 PPM which is far less than the absolute sample rate errors that were measured above.  So this periodic sample dropping glitch is not a significant source of that particular rate error.  In fact the PPM error is in the opposite direction. 

Update May 27 2008: This periodic glitch problem was verified to also exist with the Linux 2.6.18 kernel on a computer with a completely different USB chipset and CPU architecture (PPC).

Update June 17 2008: The sample rate estimation feature of the Input Device and Output Device windows were used to a measure that the UA-25 input side was roughly 0.14 PPM lower than the output side.  So the periodic glitch appears to be an ADC clocking problem.

Update June 30 2008: An accurate and known good record and playback device (RME Multiface II) on a different computer was used to test the Edirol's input and output channels.  The periodic glitch was seen on the Edirol UA-25's input ADC while the output DAC (tone generation) side was clean.  This experiment matches the result from the sample rate estimation test done on June 17th.

Update June 23 2011: The periodic glitch problem is not present when baudline runs on Mac OS X 10.6.7 (Snow Leopard).  This test result suggests that the dropped sample glitch is not hardware related but is caused by the Linux USB audio driver in the kernel.  I need to run this test again with a new Linux kernel and see if this esoteric USB audio bug has been fixed in the past 3 years.

A white Gaussian noise signal source was generated and captured in full duplex loopback fashion at each of the standard sample rates.  The Histogram plots below show a unique sample distribution that is dependent on sample rate.


Clean Gaussian shape with a slight gain mismatch between the left (green) and right (purple) channels.  This gain mismatch can easily be corrected by adjusting the left and right channel input SENS knobs.  The input gain controls were set to their minimum (counter clockwise) position for these full duplex tests.

Note that the histograms for the different sample rates and the ALSA and JACK test runs looked identical.

Channel Delay
A sine wave signal was generated and captured in full duplex loopback mode.  The time domain response was observed with the Waveform window where the green curve represents the left channel and the purple curve represents the right channel. 

The Edirol UA-25 has a zero sample inter channel delay.

The 44100 and 48000 native rates had zero loop error which means that the input and output convertor clocks are locked.  The 195 PPM absolute sample rate error seems a little high for a semi-pro audio device.

The UA-25's distortion measurements are good but not spectactular.  There is zero inter channel delay and good quantization shape.  The UA-25 signal path is clean and free from any major artifacts.

While minor, the existence of the periodic sample dropping glitch is not good and the wandering ENOB distortion performance is unsettling.  The sample dropping can be heard with sine wave test signals but it likely is inaudible with real world signals.  The ENOB wandering manifests itself and changing of the noise floor and modification of the distortion harmonic products.  So both of these artifacts are not as detrimental as they first appear. 

Good distortion measurements.  A step up from in-the-box sound cards but still far short of the theoretical 16-bit limit. 

The UA-25 has some minor quirks and despite the periodic sample dropping glitch it comparatively but it is a very good performer.  The UA-25 and it is an especially handy device with 24 bit samples, half duplex 96kHz sampling operation, optical SPDIF in/out, digital loopback option, balanced inputs, phantom powered XLR mic inputs, headphone jack, MIDI in/out, and multiple knobs and buttons.

Copyright © 2008 - group - blog - site map