This site is sponsored by FragrantSword
What a fine industry to work in! Just when we had developed a good understanding of how to design audio equipment using thermionic tubes, some bright spark decided to throw it all away and change over to transistors. As soon as we had found out how to make transistors work to the same performance characteristics as thermionic tubes, the semiconductor industry developed cost effective methods to package hundreds of transistors into a single chip, so we had to start over again. Now we are told to forget all about analogue electronics, because it is more convenient to do our signal processing in the digital domain. This means that since I started designing systems with tube electronics, I have been involved with re-inventing the wheel four times now. But I'm not really complaining, because going back over the basic principles for the fourth time reminds one that electronics must obey the laws of physics - whatever Marketing Executives may like to imagine.
As one would expect, there have been a large number of changes in the professional audio industry during the past 40 years. Of course, back in the 1960's, almost all professional audio equipment relied on circuits that used thermionic tubes. One of the less obvious differences from those days, was that much of the equipment used by members of the audio profession was also designed and built by themselves. In the UK, that meant that if you decided to set up a recording studio, you needed to be able to design and build your own mixer. The well known recording companies built their own: EMI, Decca, Pye, etc., as did the BBC. Note that the live sound and commercial radio parts of the industry did not yet exist. And there was another important difference from today - designers of professional audio equipment spared no expense in the search for the highest possible audio quality. No audio professional went looking for low cost bits of kit, and there was a general expectation that high performance electronics would cost more than the equipment one used at home.
By 1963, the expansion of the popular music culture was driving dramatic changes right across the UK recording industry. Artists demanded to use the newly available multi-track tape recorders, which led to the need for increased recording session time, which in turn created the requirement for extra studio space. Very soon, independent recording studios started to appear: Lansdowne; Advision; De Lane Lee; Olympic. At about the same time Pirate Radio stations began broadcasting from boats moored just outside the 3 mile limit, and the first "Rock Musicals" appeared on Broadway and the West End. Recording engineers invented and developed a wide range of brand new techniques: multi-track and stereo recording; echo, double-tracking, chorus, phasing, and flanging - effects which one can now buy in a box - to go.
To cope with all this - a whole new range of audio engineering manufacturers popped into existence: Sound Techniques; Neve; CADAC; Helios (in the UK); Spectrasonics; Quad 8; MCI; Harrison (in the US). Almost without exception, the newly formed independent audio manufacturing industry embraced the relatively new transistor as the basis for their circuit designs. Physically smaller, more convenient and much easier to use, the transistor proved to be difficult to integrate into the high performance designs that were equivalent to the best tube equipment of the day. The more obvious reasons being: very much reduced headroom; limited bandwidth; poor transient response. Never-the-less, by the early '70's, semi-conductor electronics had been tamed and the search for high performance audio was well and truly in top-gear once more (overdrive still being sought today).
Into this heady mixture of science and art came improvements in loudspeaker design, and a better understanding of room acoustics. This led the artistic branch (balance engineers and producers) to demand higher definition electronics in the mixer/processing departments. Luckily, most of us involved in designing audio equipment in those far off days had learned much of our trade on-the-job, so we were pretty much used to juggling the various parameters that affected the performance qualities of our circuits to obtain the engineering compromises necessary to solve the artistic requirements. Quite often, the work bench was only a few meters away from the studio or control room, so one was able to get confirmation of how good a new circuit sounded almost instantly. We found that higher definition at low frequencies was possible by improving the transient response and phase response of the circuits. Better high frequency definition was obtained by increasing the bandwidth of the input amplifier, which was itself involved with the phase response characteristics, as we will see.
It turned out that understanding the vagaries of the phase response of a circuit or system was, and still is, essential for evaluating how good the design really is. Unfortunately, a graphical plot of a system's phase response does not look very flattering, as can be seen from the accompanying figure, so marketing executives have long since banned such things from publication (along with polar plots and the power response diagrams for loudspeakers).
The phase response of a system is important, because the human auditory system is extremely sensitive to changes in phase. Detecting and processing the results of the phase relationship of a sound source allows us to locate its direction and position with extraordinary precision - even if the sound comes from behind us. Think of the situation where someone calls your name - in a room or in the street. The sound reaches first one ear, and a few microseconds later - the other ear. The brain then calculates the angle represented by the difference in the arrival times of the sound at each ear, and we know precisely in which direction to turn in order to find the sound source. In fact, most of us are able to work out how far away to look for the sound source - based on our experience of the natural loudness of the various sounds that we know about or regularly work with - stored in our memory.
Now, the ability to respond to changes in phase evolved mainly for the preservation of life - in the days when humans roamed the earth as hunters and gatherers. Since then, many of us have adapted the same mechanism to help us appreciate the complexities of music and drama. A small number of audio professionals have become even more specialized in ear resolution, by training their auditory system to recognise extremely small phase changes at either end of the audio spectrum. For their efforts, they are often awarded the accolade: Golden Ear Brigade. While this is often used as an insult, it is usually found that the person voicing the phrase in this manner has the hearing acuity of dead slug.
As can be seen from the phase response diagram, the rate-of-change of phase is greatest at low and high frequencies. Listening tests reveal that humans are particularly sensitive to changes in phase when the rate-of-change of phase is excessive. For instance, we are easily able to detect changes in the timbre of low frequency sounds if the rate-of-change of phase in the reproduction system is more than about +3 degrees per octave. So, what is it that controls the low frequency phase response in electronics?
When we consider the frequency domain, it is immediately obvious that the input circuit for any electronic device is essentially a high-pass filter. The low frequency response is therefore determined by the value of the input capacitor Ci, for a given value of input impedance Zi. For a modern transformer-less microphone amplifier, it is generally agreed that a reasonable value for Zi is around 1k ohms. This is based on engineering experience that tells us that the input impedance (or load) should be at least 5 times more than the source impedance. So, an input impedance of 1k is based on the understanding that the source impedance of the microphone is about 200 ohms. A simple calculation then tells us that the value for Ci only needs to be about 0.5uF, in order to get a "-3dB down" point at 20Hz. It is not until we consider the phase response at 20Hz that we realize that such a low value for Ci will be a sad mistake. If the frequency response is -3dBu at 20Hz, then the phase response at the same frequency will be in the region of +40 degrees. The rate-of-change-of-phase is excessive and a listener who is used to working with sound sources that develop complex low frequency harmonics, will complain that the reproduced sound is not accurate. A quick look at most allegedly high performance microphone or line amplifier circuits will reveal values for Ci of at least 100uF, and sometimes an even higher value. High value capacitors are expensive, but the designer is not throwing money away, if he is interested in high performance low frequency reproduction. Large value input capacitors relate to an extended low frequency response and low rate-of-change phase response.
At high frequencies, the phase response of a circuit or system turns negative, because the combined circuit elements behave as a series of parallel connected low-pass filters. The rate-of-change of phase begins to increase rather rapidly above a frequency that is equivalent to 1/10th of the turn-over frequency of the combined filter network. Thus, the behaviour of the high frequency phase response changes with the overall bandwidth of the circuit or system. Our natural ability to detect changes in phase allows us to resolve quite subtle changes in the overall bandwidth of a system.
In the early 1970's, I was involved in a series of tests to try and determine the minimum bandwidth required for an input amplifier. Using a calibrated microphone and a spectrum analyser, we recorded the overall harmonic response of a number of different musical instruments. It was immediately clear that many acoustic instruments produced enormous amounts of harmonic energy well beyond the accepted audio bandwidth of 20Hz to 20kHz. Luckily, there is an old Polaroid photograph left over from the session, which will give you some idea of what we observed on the Spectrum Analyser screen. This photographed was taken as a percussionist gently rattled a tambourine.
Six special microphone amplifiers were constructed to allow the input bandwidth to be switched from wide-open (100kHz) to 40kHz in 20kHz steps. A recording session was organised to record a simple jazz trio - piano, bass and drums. The listening tests revealed that everyone in the control room could hear when the bandwidth of the amplifiers on the percussion microphones was reduced from 100kHz to 80kHz. Reducing the bandwidth to 60kHz further modified the percussion sounds, but also changed the timbre of the piano. When the bandwidth of all the microphone amplifiers was limited to 40kHz, the reproduced sound on all three instruments was heavily modified. Of course we had the advantage of being able to instantly compare the original sound sources with the electronic reproduction, by walking between the control room and studio. Doing A/B comparisons is the only way to be sure that sound "A" is different to sound "B".
Back in the lab - looking at the frequency domain on its own did not indicate any sort of electronic anomaly. But, when we checked the phase response, it was clear what was happening. As the overall bandwidth was reduced, the rate-of-change in the phase response increased, and the frequency at which the phase response turned negative (1/10th of the upper band-pass frequency) fell as the overall bandwidth was reduced. In our case fs/10 - decreased from 10kHz (when the overall bandwidth was 100kHz) to 4kHz (when the overall bandwidth was reduced to 40kHz). Note that fs/10 was always in a part of the audio spectrum that a human with normal hearing could easily resolve.
Low frequency phase response is still a challenge to most of the A/D converters currently available. Many chip designs require - or have a built-in high pass filter network before the converter. This is required to minimise the effects of the DC off-set voltage between the analogue input stage and the converter, which causes a number of anomalies in the converted digital data. As is often the case in engineering - you solve one problem and get another one in its place. For many of us, the heavily modified phase response caused by the high pass filter is at least as bad as the errors produced in the digital output data that occurs without the filter in place. Listening tests on early A/D converters always produced the reaction that there was the lack of low frequency extension compared to the analogue input circuits that we were used to. Well, as soon as one saw the results of the phase response tests, it was obvious why low frequency extension was lacking in so many A/D circuits. The effect of the high pass filter on the frequency response at low frequencies appears to be negligible - being typically -0.4dBu at 20Hz - but this relates to a phase response of more than +30 degrees at 20Hz! Compared to the high performance analogue circuits that I got used to working with for thirty years (which have a phase response of +3 degrees at 20Hz), +30 degrees of phase-shift at 20 Hz makes a bass drum sound very thin indeed.
It turns out that many manufacturers of digital equipment do not test for phase response. It is also possible that not many of them do listening tests either. Admittedly, checking the phase response is difficult and time consuming for digital circuits, but it never occurred to me not to do them.
At high frequencies, the phase response of an ADC is controlled by the sampling frequency used. The higher the sampling frequency - the lower is the rate-of-change-of-phase at high frequencies. The reason for this is because the physics is the same as for analogue input circuits: as the sampling rate (bandwidth) increases, the turnover frequency at which the phase response goes negative (fs/10) increases with the sampling rate. But there are many people who doubt that there is important musical information below 20Hz and beyond 20kHz. My answer to that is: have a look at the amazing research done by James Boyk of Trinity College, Western Australia. Of particular interest to me is: "There's life above 20 kilohertz! A survey of musical instrument spectra to 102.4 kHz" - www.cco.caltech.edu/~boyk/spectra/spectra.htm .
The fact that human listeners can respond to the effect of frequencies below 20Hz and above 20kHz should not be doubted. This is because the human auditory system is extremely sensitive to the rate-of-change-of-phase. It has also been demonstrated that frequencies are generated by acoustic or electronic sources that are well outside of the generally excepted audio pass band of 20 Hz to 20kHz.
When we are designing or specifying high performance audio equipment, we need to be aware that a number of different parameters gang together to effect the overall outcome of the system response. Focussing on a single parameter will not be enough to explain what is actually going on, because all of the different parameters interact with each other, due to the fact that they are inextricably linked by basic physics. Note that so far, we have only discussed two parameters - the response to frequency and phase! High performance audio engineering requires that the frequency response of our input stages must have a bandwidth wide enough to minimise the rate-of-change of phase at the extreme ends of the audio spectrum. Unfortunately, checking the phase response of digital-audio systems is difficult and time consuming, but that is what we need to do if we are interested in accurate sound reproduction. Don't be surprised if you can hear the difference between 48kHz and 96kHz sampling converters, but be very worried if you cannot!
email Tony: twaudio@hotmail.com ; cell 'phone: +44 (0)7932 863670; postal address: 24 Knoll Rise, Luton, LU2 7JA, UK.
copyright © Tony Waldron 2005
This site is sponsored by FragrantSword