Optimal REW settings for measuring amplifiers

[reallystate]

Hello everyone,

I'm starting to use REW to measure the performance of my headphone amplifiers and I have several questions about the optimal configuration. I've done some reading, but I would appreciate some expert advice to ensure my measurements are accurate. My main questions are:

• Should I measure at the maximum possible sample rate, or should I limit it to 48 kHz?
• Should I use low-pass and high-pass filters? If so, at what frequency should I set them?
• What FFT value and window type should I choose? Should these values be changed for different types of tests (THD vs level, IMD vs level, etc.)?
• What sweep length value should I choose when measuring the frequency response of a device, since this choice affects the phase measurement result?
• What buffer size should I set in the sound card settings section?
• Should I use the Overlap option?
  
  Thank you in advance for your guidance.

 

[John Mulcahy]

Should I measure at the maximum possible sample rate, or should I limit it to 48 kHz?

Generally best measuring at the lowest rate that covers the bandwidth you are interested in.

Should I use low-pass and high-pass filters?

Can't think why you would.

What FFT value and window type should I choose?

For RTA measurements longer FFTs would give greater frequency resolution and lower noise floor. 64k is generally sufficient for many purposes. Blackman-Harris 7 is a good window choice for most measurements, though you could use a Rectangular window if your input and output are on the same interface.

What sweep length value should I choose when measuring the frequency response of a device, since this choice affects the phase measurement result?

No, sweep length doesn't affect phase. Longer sweeps give greater signal-to-noise and a lower distortion noise floor.

What buffer size should I set in the sound card settings section?

16k is usually sufficient.

Should I use the Overlap option?

Using maximum overlap will produce faster RTA display updates, each update needs to wait for sufficient samples to be acquired.

 

[reallystate]

Thank you for your responses.

Can't think why you would.

When i use low-pass and high-pass filters, the noise is lower

Generally best measuring at the lowest rate that covers the bandwidth you are interested in.

I don't know what sampling frequency is worth looking at for measuring amplifiers. At a sampling frequency of 48 kHz, the noise is lower than at 192 kHz. And if you also enable filters at 192 kHz, the noise will be even lower. On the other hand, the use of filters can negatively affect the calculation of the distortion level.

No, sweep length doesn't affect phase

Is this a bug or a feature?

 

[John Mulcahy]

The distortion limits for most published specifications are 20 Hz to 20 kHz. Sample rate will have an effect if your limits exceed half the sample rate.

For sweep length differences attach the mdat file with the measurements rather than images.

 

[reallystate]

For sweep length differences attach the mdat file with the measurements rather than images.

 

[reallystate]

Sample rate will have an effect if your limits exceed half the sample rate.

When I measure a certain device at -20 dBFS using 192 kHz with 20-20k Hz filters, the noise level is -81.5 dBr, but when I use 48 kHz with 20-20k Hz filters, the noise level is -75.9 dBr. This can be seen in the screenshots above. What sampling rate should I use in this case?

 

[John Mulcahy]

Looks like an issue with clock rate differences between the input and output devices. If you add a loopback connection as a timing reference REW can use it to correct the clock difference.

 

[John Mulcahy]

Interface behaviour can also change with sample rate, best check that. For headphone amps 48k should be fine.

 

[reallystate]

When conducting THD vs level and THD vs frequency, the noise floor level is measured and the indicators differ in both cases. Can we assume that the results of the second test better reflect the real situation? What frequency is Noise floor measured at when performing a THD vs level test? According to the generator readings, it is at 1 Hz. But judging by the readings on the RTA screen, it is measured at different frequencies.

 

[John Mulcahy]

The noise floor figure for stepped level THD measurements is the maximum noise level within a one octave span centred on the the stimulus frequency. THD vs frequency shows the noise floor spectrum.

 

[reallystate]

Can you increase the amount of space allocated for the measurement name display below the graph? There's still plenty of space to the right of the graph names in the exported image, but the names still don't fit.

 

[John Mulcahy]

It's a little difficult to manage. All the legend entries for the various graphs need to keep the same width. I can adapt the width to the measurement names, but it would have to be to suit the longest name of the currently loaded measurements otherwise the legend height could change when moving between individual measurements. I'll look at that for the next build.

 

[reallystate]

Is it possible to implement stepped measurements of devices that don't support REW? For example, I want to generate an audio file, transfer it to an audio player, and then play it while connected to an audio interface, thereby measuring it (THD etc. vs level, THD vs Frequency).

 

[John Mulcahy]

No, making that work reliably (or at all) would be an awful lot of work on both the signal generation and capture analysis sides for a very niche use case.

 

[akl]

Measuring amplifiers, especially headphone.

Important is the LOAD impedance, I use 32R _and_ 150R ,
sometimes also 16R and 600R.

I usually make a fast test stating the max. usable level . I use the SCOPE at 1 kHz and look at the clipping point.

You have to have a RMS Voltage Meter for the basic calibration. A cheap averaging DVM or simple needle VoltMeter will NOT do.

You also have to use a precise DA -- DUT ( Device under Test ) -- AD to get REW working, the distortion of the DA AD will add to the DUT .

I use RME ( www.rme-audio.de ), high but fair price units with precise switchable gain .
The 75 Ohm OUTput impedance is quite usefull, many units with a semi-floating output.
10 k Ohm INput impedance is also suitable with todays equipment.

But evaluating the main distortion performance I use only TDFD (TWIN tone) measurements. Bass 41 89 Hz measuring the low end, and the 13k5 20 k stimulus for the high end. The Reference Level is the Level of the single stimulus + 6dB.

I always use decibel. Especially for headphone stating the power dBm . With dBm you have always to state the impedance.
( Also usefull for power LS-amps ).

Using the decibel is easier to get a feeling about performance and easier to compare performance and correlates quite good with the ear .

I also state the noise ( UNweighted 22-22k ) , 'A' and CCIR-468 QP with Reference to 1 mW ( 0 dBm ).


To judge Distortion at the HIGH frequency edge harmonic will _NOT_ do.
THD is not possible at high slew rates in standard digtal system ( CD MP3 ).
The TDFD akl ( 13k5 20k ) showes a high SLEW rate.

You have to use a TWIN Tone signal . TDFD has most advantages.

BASS , using TDFD 41 89 gives high sensitivity and reveals problems a harmonic distortion analysis will not.

REW is to my knowledge the only Methode to use TDFD TotalDifferenceFrequencyDistortion with comfort and precision.

You may also name TDFD Thiele DifferenceTone Distortion,

You will know Thiele ( and Small ) from the T&S Loudspeaker Parameters.
Thiele revealed this methode to controll NONlinearities down under. TDFD still deserving wider recognition, though registered at IEC .


Since you measure at the digital domain, you have to know
the dBFS value ( digital FullScale) in reference to
0 dBm 15R and 150R .
mind: 0 dBm @ 600R ( 1 mW ) == 0 dBu == -2.22 dBV

1 V == +15.0 dBm @ 32R ( 31 mW ) == 0 dBV ( +2.22 dBu )
== +14.8 dBm @ 33R ( 30 mW )
== +8.2 dBm @ 150R ( 6.7 mW )

32 Ohm = 16R + 16R = 33R // 1k
(32 Ohm is the standard but you may use 33 Ohm or build a 32 Ohm Resistor)

For some amateurs ist may be the eaysiest way to use standard size resitors,
each channel
6x 100 Ohm --> 150 Ohm
7x 220 Ohm --> 32 Ohm -0.16 dB
only super headphone amps will overload these loads.

A last word (in my humble opinion):
Zero Impedance headhone amplifiers are Myth
symmetric outputs for headphones are Myth
BUT:
Connectors with separate 4-pole are important,
especially at 32R and 16R (!)
minimizing distortion and CrossTalk

-akl- 51109