Notes from the Test Bench
By Bruce Hofer, Chairman & Co-Founder, Audio Precision
Having worked in test and measurement for most of my life, I take specifications very seriously.
I view specs as a promise that the product described meets or exceeds a certain level of performance under a variety of environmental conditions.
AP has a reputation for being conservative in our specs because we take the obligation we’ve undertaken to our customers extremely seriously. Our instruments have to perform at or above specification in hot and humid countries, cold rooms, in labs below sea level, and in radio stations on the Tibetan plateau. Because we spec for the worst case, not the best possible, our instruments always perform as promised or better.
The quality of specs out in the real world varies almost as widely as the products they characterize. Some devices are highly specified with clearly stated conditions and measurement parameters—others print claims that are so meaningless it’s almost funny.
Being loose with specs can look good on the upside, and I can almost promise that every engineer will feel pressure at some point in their career to specify better performance, be it from marketing looking for a competitive advantage to tout, or from upper management, who needs to meet an engineering milestone in any way they can. Some engineers will be vague, giving a “typical” spec that has no defined (or reproducible) conditions. Others will try to compromise with a bold banner spec, while burying the truth in footnotes.
All I will say is “Beware.” Ultimately, engineers are responsible for the specifications they publish, and if specs are discovered to be misleading, then the damage to a brand can be catastrophic.
This month, AP is releasing the first in a series of white papers on writing audio specifications properly. It is well worth reading, for both producers and consumers of audio specifications. Based on your input, we will continue the series, looking at different classes of devices.
I hope to see many of you at AES London next month—until then, enjoy the paper, and always state your conditions clearly.
Output: How to Write (and Read) Audio Specifications
Below is an excerpt from our new white paper “How to Write (and Read) Audio Specifications” by Audio Precision’s Senior Technical Writer Dave Mathew. In the full version, each of the guidelines is discussed in detail, and key specifications for two classes of audio device are examined for correct expression and form, with examples of real-world specs for comparison. You will also find more information about audio standards, and a discussion of THD vs. THD+N.
by Dave Mathew, Senior Technical Writer
Frustrated by audio specifications that are confusing? We’ve all found that getting a clear picture of a device’s performance from a data sheet is not as straightforward as you’d hope, and that comparing devices from different manufacturers is even worse.
At Audio Precision, specifications are very important to us. We make, and spec, the analyzers audio engineers use to spec their own designs, assuring quality in their manufacturing facilities and providing the performance benchmarks that their sales force can use in positioning the product relative to the competition.
Recommendations for writing specifications
We consider these guidelines as we write our own specifications, and think about them when we read specifications from other manufacturers.
In this excerpt, we’ll look at the most important of these in more detail. It’s fourth in the list:
State ranges and conditions
Stating ranges and conditions is absolutely essential. Many specifications mean nothing if these parameters are not provided. Yet, most manufactures do not include this information in their specifications, or what they do include is incomplete. Are they saving ink?
Every level measurement is made at some frequency or range of frequencies. Every frequency measurement is made at some level or range of levels. Be very clear as to the level and frequency of the stimulus signal, and to the level range, bandwidth and filtering in the analyzer.
External conditions and device configuration conditions, too, will affect device performance. What is the mains voltage? The ambient temperature? Is the device volume control set to low, medium or high? Are all channels being driven, or only the channel being measured?
If there isn’t room for the conditions in the brochure, there certainly is in the data sheet. Annotate specifications when more detail is required. Use footnotes or endnotes.
Here is an egregious but not uncommon example: An expensive audiophile amplifier states its THD+N this way:
What does that mean? There is no level or bandpass filtering mentioned. It’s like saying “I’m taller.” THAN WHAT?
The table below shows the THD+N results for one channel of a home theater receiver, made under varying conditions. Rated power (100 %) is 94 W into 4 Ω. It’s clear that THD+N stated with no conditions is meaningless.
THD+N measurements of a home theatre receiver made under varying conditions.
Establish that the measurement used for a spec is not a “best result” of a “golden unit” but is typical of the device, achievable by any properly adjusted unit.
It’s always possible to tweak a device (by carefully selecting components, for example) so that it will surpass the typical performance of the devices. Do not use such results for specifications.
It is also possible to find a “sweet spot” in a characteristic, and report that as a performance specification. That is misleading, even if the specification specifically mentions the sweet spot.
For example, the following graph shows the residual THD+N versus amplitude for one of our analyzer systems. For audio analyzers, this is a key specification, one we would like to tout.
Audio analyzer THD+N vs. amplitude, system performance (generator and analyzer). The sharp inflections in the trace are artifacts of range switching (a characteristic of all audio analyzers), and the rise at the left is the inevitable approach of the noise floor. Red line equals -112 dB.
We claim ≤ –112 dB residual THD+N in a 22 kHz bandwidth. You can see that for most amplitudes above 400 mV, the THD+N is well below this (the red line). If you are working at a 2 V amplitude, it is about –116 dB. We find, at least in our end of the business, that it is important not to mislead our customers.
How to Write (and Read) Audio Specifications (the full 8-page version of this article).
Sound Advice: AP Knowledge Base
Measuring Non-flat Devices With Equalized Multitones
The new APx Multitone EQ Utility, written by AP's Director of Technical Support Joe Begin, creates multitone stimulus files that are equalized, rather than being flat as usual. Read on to see why that's useful, and how the utility works.
When testing certain types of devices or systems that are known to have a non-flat frequency response (e.g. phonograph preamplifiers and other pre/post equalized signal chains), it is often desirable to inversely equalize the audio test stimulus signal, so that frequency response measurement of an ideal device appears as a flat curve.
The flat curve makes it easy to visualize and quantify defects in a device’s equalization circuits. In many cases, equalizing the stimulus signal is superior to the alternative of mathematically correcting the results, as it keeps the stimulus signal within the device’s normal operating range at each frequency, thereby preventing errors due to overload. In the case of a phonograph preamp, for example, a 20 Hz input stimulus may clip at only 30 mV whereas a 20 kHz stimulus might not clip until it exceeds 500 mV
Using the APx Multitone EQ Utility
The utility creates an equalized multitone stimulus signal, with the inverse response curve calculated either by measuring an existing device with an APx analyzer, or by importing a Microsoft Excel Worksheet file with frequency and level data.
Figure 1 APx Multitone EQ Utility using "Measure with APx" as the data source.
When “Measure with APx” is selected as the data source, the APx signal path and measurement boxes default to “Signal Path1” and “Multitone Analyzer”. Note that the active APx project must contain a multitone measurement within the specified signal path—if one is not present, then it should be added to the measurement navigator.
Figure 2 Measured frequency response of the reference device (bottom), with the derived inverse equalized stimulus (top).
The reference device is now connected to the analyzer, after correctly setting the signal path connections and generator level. Any subsequent device tested using the equalized multitone that we are about to create, will measure as flat if exactly matches the reference. Click the "Measure Frequency Response" button to perform a multitone measurement and collect the frequency response data.
Figure 3 Retesting the reference device or a perfect clone, using the equalized multitone, will result in a flat frequency response curve.
Now that the data is collected, it's time to create the inversely equalized multitone. Select the desired sample rate and bit depth, assign a filename, and choose if you want the file automatically uploaded to the APx generator upon creation (recommended). With the equalized multitone waveform loaded into the generator of the Multitone Analyzer, retesting with the same device should result in a flat frequency response.
Figure 4 APx Multitone EQ Utility using "Read from APx exported file" as the data source.
If “Read from APx exported file” is selected as the data source, you can click the “Import…” button to specify the desired Microsoft Excel workbook file. The workbook data must be in a specific format—to make a template, run the APx Multitone Analyzer measurement and then choose "File | Export Graph Data" to create an Excel workbook. Then, open the workbook, go to the Relative Level worksheet, and replace the data in the Y column(s) with the desired numbers.
Figure 5 Microsoft Excel worksheet with exported graph data.
Typical reasons to edit the data are to correct for errors in the reference device's response, or to use a curve that is a published standard. If the frequency you actually want to modify is not listed in the worksheet, then you will need to interpolate accordingly. The frequencies included in the APx 32-tone multitone are essentially the ISO preferred 1/3-octave frequencies from 20 Hz to 20 kHz. Note that if the Relative Level source data contains only one channel, then both the left and right channels of the stereo .wav file will be equalized using the Ch1 data.
APx Multitone EQ Utility
Test Results: AP News & Events
©2010 Audio Precision Inc.