Sonic Signatures: The Art and the Science
by, 05-24-2012 at 03:29 AM (3835 Views)
I believe every piece of audio equipment has its own sonic signature. E.g. CD transports, cartridges, tone arms, turntables, preamps, amps, cross overs, speakers, interconnects, basically every component, part, and wire in (and around, e.g. power supplies) the audio signal path will have its own sonic signature, whether designed or not. Technically, a sonic signature is called a transfer function, but we will get to that shortly. I also believe there is a direct correlation between what we hear and what we measure. Again, related to transfer function.
Let’s go back in history and see what the state of the art in audio design was say in 1953, almost 60 years ago. The audio designers “bible” at that time was The Radiotron Designer’s Handbook: Radiotron Designer's Hanbook on CD-ROM - Circuit Cellar, Inc. .
The Radiotron Designers handbook has been written as a comprehensive, self-explanatory reference handbook for the benefit of all who have interest in the design and application of audio amplifiers. The book was designed to be as self-contained as possible.
What is interesting, even in 1953, there was a collection of state of the art valve amplification designs, including amongst other notable classics was a single ended triode Class A design, regarded among many, even today as the most musical sounding amplifier design. At 1400 pages, the book contains everything an aspiring amplifier designer needs to design and build a high fidelity amplifier.
Back in 1953, the definition for high fidelity was, “True fidelity is perfect reproduction of the original. In reality, true fidelity can only be regarded as an ideal to aim for.” I believe that definition still holds true today. As does many other principles, designs, mathematical theories, measurement standards, etc., all encompassed in this classic book on audio amplifier design.
For example, most of the types of audio distortions were already well known in 1953. Not only were they classified by type, but also how to measure each one against a standard, and how to interpret the results by characterising the sound quality using subjective terms.
Put aside for a moment the revolution that is digital audio, and we will see that analog audio design has not changed much from the state of the art since 1953. Of course, there has been many “material” changes that have dramatically increased the fidelity, but there have only been a few new “design” innovations.
For example, back in 1953, the fundamental class or type of amplifier designs were mostly figured out and well documented. Electronic amplifier - Wikipedia, the free encyclopedia Of course, new innovations produce new amplifier classes. However, each amplifier class has its own sonic signature. Technically, this sonic signature is called a transfer function.
An excellent definition of a transfer function, as related to audio design, can be found in Bob McCarthey’s most awesome book, Amazon.com: Sound Systems: Design and Optimization, Second Edition: Modern Techniques and Tools for Sound System Design and Alignment (9780240521565): Bob McCarthy: Books
“The response of a device from input to output is its transfer function. The process of comparison measurement between input and output is termed transfer function measurement. This could be a passive device, such as a cable, attenuator or filter, or an active analog or digital circuit. The transfer function of a device is obtained by comparing its input signal to its output signal.”
“The transfer function of a hypothetical perfect transmission system would be zero. Zero change in level, zero delay, and zero noise at all frequencies. Any device that passes signal will have some deviation in transfer level and transfer time and will add some noise in the process. Transfer function measurement can detect these changes and has a variety of ways to display them for our analysis.”
Wikipedia’s definition of transfer function is more formal, “A transfer function is a mathematical representation, in terms of spatial or temporal frequency, of the relation between the input and output of a linear time-invariant system.” Whew, have a look at the math on that page. What is interesting to note is that this has all been figured out long ago and part of the science of audio engineering.
What’s my point? If we go back to the class or type of amplifier design in my example, these are the corresponding transfer functions of Class A, AB, and B amplifiers:
Note, because of these transfer functions, and other inherent characteristics (or properties) of different classes or types of amplifiers, each class will have its own unique sonic signature. Personally, I like the sonic signature of Class A amplifiers. Subjectively speaking, excellent wide-band damping factor (i.e. control) especially into highly reactive loads:
Wide bandwith response (100 KHz) with low distortion:
There is a direct correlation between what we hear and measure based on these transfer functions. Not only can the transfer functions be fully measured, but can also be represented mathematically (in a software modeling tool) so the designer can model in software, the sonic signature, run tests to verify the design. The designer can then implement the design that is a transformation into a physical device (or a software runtime). Rerun the measurements and verify the implementation meets the design. Been a reality for quite a while.
The art is in how a particular designer uses the science to create sonic signatures (i.e. transfer functions).
Let’s take an interesting example. Remember back in 1985 the Stereophile Bob Carver challenge? You can read about it here The Carver Challenge | Stereophile.com and I attached the PDF. It is a fascinating read.
The point is that Bob was able to modify and match the transfer function of his amplifier to a well-known audiophile amplifier. How did he do this? He used a null test technique to adjust the transfer function in his amp to match the transfer function of the reference amp. The deeper the null, the more his amp will sound like the reference amp. In fact to the point where the Stereophile folks admitted defeat.
Therefore a person can design, model, and replicate anyone’s transfer function (e.g. sonic signature). This actually has been going on a long time in the world of pro audio.
I invite you to watch this 5 minute video from one hardware turned software manufacture. I used their real physical hardware (e.g. Studer, 1176LM), now these are all emulated in software, including the most sought after characteristic, their sonic signatures:
UAD-2 Powered Plug-Ins Platform | Digital Audio Products and Plug-Ins | Universal Audio
Did you happen to notice the digital representation of the Studer A-800 analog tape recorder? I used to work with the physical machines and I know it’s sonic signature only too well. To “hear” that characteristic sound emulated in software is mind blowing.
Consider the Universal Audio 1176 limiter, the hardware version here: 2-1176 Twin Vintage Limiting Amplifier | Universal Audio and the software version here: 1176 Classic Limiter Plug-In Collection If you read the blurbs, you will understand exactly what I mean by sonic signature. And the fact that they have both a real physical version and a virtual emulated software version, in which I can assure you have been nulled tested to death to ensure the units have the exact same sonic signatures.
The other reason that this is relevant is that almost every piece of rock, pop, jazz, blues, and other multi-tracked, and even live to two track, mastered, pressed, etc., has passed through this device. Everyone listens to the sound, but can you hear it? I have spent so much time with this device, as others have in the recording industry, that I hear it on almost every piece of music I listen to. In fact, for every tune I listen to, I mentally turn the dials on the ol 1176 to the settings I think I am hearing. That’s how much I hear it. I made a comment about this that I can track the limiter on Nora Jones album and if you read the blurb on the 1176 hardware version, you will see they reference Nora Jones using the device. I try and not let this interfere with my enjoyment of the music.
So can measurements measure everything about audio? Of course. The above proves it. How else can you emulate the sonic signature of a piece of audio hardware in software. Transfer function testing is used all the time (including null testing) to make this happen. Btw, if you paid close attention to the video, check out the amount of test gear.
For the software, The Audio Programming Book is a great place to start. The Audio Programming Book - The MIT Press This has already been figured out and commercialized in the pro audio industry. In addition to the Universal Audio software plugins, there are literally thousands of already developed and commercially available plugins in the largest plugin database in the world. KVR audio KVR: All Plug-ins / Hosts / Apps On One Page
Null testing is one proven technique to measure sonic signatures (i.e. transfer functions). It is backed up by science and math and been utilized by designers for decades. The 1954 Radiotron book, The Universal Audio example, as with the Bob Carver example, and even examples on this very site prove it’s real.
If you are interested in how this really works, Amazon.com: Sound Systems: Design and Optimization, Second Edition: Modern Techniques and Tools for Sound System Design and Alignment (9780240521565): Bob McCarthy: Books book is an excellent one. Of course, there are several other references, textbooks, DSP sites, etc., that walk you through the science and practice.
You can try the null test yourself, like esldude did in his post here. I have performed many null tests in my electronics and software programming career. Nowadays it is really easy as we have access to free tools like Audacity and Audio DiffMaker (or any Digital Audio Workstation for that matter) that have incredible dynamic range and measurement sensitivity.
Here is a null test that anyone can perform with this free software and can easily repeat the same test and get the same results I did. In fact you can listen and hear the differences yourself as I attached the difference files to this post. Take any WAV file and convert it to an MP3 and either using Audacity or Audio DiffMaker, you can compare the difference between a lossless file format and a lossy format.
I preformed these tests using both Audacity and Audio DiffMaker and got similar results. Note I used the LAME MP3 encoder at 320 Kbps. Listen to the files, that is the absolute signal difference between WAV and MP3. There is nothing else. That's the entire transfer function.
The fact that there is a difference is no surprise to me and that I don’t get a deep null is no surprise either. In fact, you should not get a deep null, you should get roughly a null of in the -40 to -50 dB range as that is the entire signal difference (or transfer function) between lossless and lossy file formats.
As a side note, remember that a transfer function, by definition, includes everything in the amplitude, frequency, and time domains. Not only can we measure it, but designers routinely model these transfer functions (in software) to produce certain sonic signature as evidenced by the video above. Null testing captures the absolute signal difference between the device under test and the reference. The science (read: math) proves this just like Ohms law can be proven.
I am not saying performing null testing is easy. One sample difference can offest the results, so precision is required. Further, there are so many interacting components within even the simplistic audio system, makes it non-intuitve to figure out. Setting up null test experiments takes a lot of understanding of how things work. For example, sometimes people hear huge differences in speaker cables. I have heard it myself. However, upon closer analysis, and the particular amplifier I was using at the time, adding a little extra cable capacitance (measured!) was just enough to destabilize the amplifier into ultrasonic oscillations, caused by poor amp design (and a sagging power supply). So was it really the cable that sounded so different? No in this case, the cable caused the amplifier to oscillate ultrasonics that produced a very harsh sound. Of course, since the cable was swapped, the immediate conclusion is, oh it must be the cable. Well, no, not really.
The much more interesting question, is the WAV vs MP3 difference audible in relation to the program level? In fact, that is the question isn’t it. At what level of signal difference can our ears (read: brain) perceive the difference? If you took the time to read the Stereophile Bob Carver challenge, you will see that at -70 dB null, the difference to the listeners was inaudible. Based on my own tests, I would concur. It seems near the -50 dB range is the limit where, for me anyway, I start not hearing any difference (using ABX plugin in Foobar2000).
For me, the reason I perform transfer function tests is because I am trying to get as close to reproducing the correct timbre in my listening room as possible. As I have mentioned in other blog posts, I believe the weakest link in the audio chain is the speaker to room interface. In fact, you can measure its transfer function, both in the frequency and time domains, using excellent (free) software like REW. And just like the emulation of the studio gear above, I can emulate the sound of a really nice room by using the power of digital audio, (DSP, DRC, and Convolution), but that is the subject for another post.
Does every piece of gear have a sonic signature. As barrows says, everything matters. You bet. At what level is it audible at seems to be the real question. Do we have the measurement capabilities to measure all aspects of audio gear? Based on the Universal Audio example above, the answer must be yes. How else can a hardware sonic signature be modeled and emulated 100% in software if you can’t measure 100% of the hardware's sonic signature (a.k.a. transfer function)?
Audio is both an art and a science. It has both subjective and objective components. So really, there is no subjective versus objective, one is simply a different view of the same thing. Remember even back in 1953, the Radiotron book they had both objective measures and subjective descriptions for those measures. We still do that, but the art and science, along with the digital audio revolution, has advanced to the stage of perhaps diminishing returns.
For example, Bob Katz and his book called, Mastering Audio, the art and the science (another awesome book that correlates what we hear with what we measure) states in the section on converters that, “All the converters mentioned here are at least A grade. The difference between an A and an A+ is extremely small, perceptible by only the most discriminating listeners and opinions vary on which is better.” You can see some of the transfer function measurements, they are literally textbook perfect – can’t get any better.
Soon we will see DAC emulation in software, where you can get your favorite DAC sonic signature :-)
Enjoy the music!