Tubes Versus Transistors - Is There an Audible Difference?*

by Russell O. Hamm

(Part 1 of 3 parts)

Go To Part 2
Go To Part 3
Back to "The Big Idea"

Sear Sound Studios, New York, N.Y.

*Presented September 14, 1972, at the 43rd Convention of the Audio Engineering Society, New York.

(Emphasis added by BK Butler as it directly pertains to Butler Audio's Thermionic Tube Amplifiers)


Engineers and musicians have long debated the question of tube sound versus transistor sound. Previous attempts to measure this difference have always assumed linear operation of the test amplifier. This conventional method of frequency response, distortion, and noise measurement has shown that no significant difference exists. This paper, however, points out that amplifiers are often severely overloaded by signal transients (THD 30%). Under this condition there is a major difference in the harmonic distortion components of the amplified signal, with tubes, transistors, and operational amplifiers separating into distinct groups.

INTRODUCTION:

As a recording engineer we become directly involved with the tube sound versus transistor sound controversy as it related to pop recording. The difference became markedly noticeable as more solid-state consoles made their appearance. Of course there are so many sound problems related to studio acoustics that electronic problems are generally considered the least of one's worries. After acoustically rebuilding several studios, however, we began to question just how much of a role acoustics played.

During one session in a studio notorious for bad sound we plugged the microphones into Ampex portable mixers instead of the regular console. The change in sound quality was nothing short of incredible. All the acoustic changes we had made in that studio never had brought about the vast improvement in the sound that a single change in electronics had. Over a period of several years we continued this rather informal investigation of the electronic sound problem. In the past, we have heard many widely varied theories that explain the problem, but no one, however, could actually measure it in meaningful terms.

PSYCHOACOUSTICS

Anyone who listens to phonograph records closely can tell that tubes sound different from transistors. Defining what this difference is, however, is a complex psychoacoustical problem. Any investigation of this admittedly subtle phenomenon must really begin with a few human observations. Some people try to point out and describe valid differences. Others just object to the entire thesis and resort to spouting opinions. It is the listener's job to sort out the facts from the fiction.

Electrical engineers, especially the ones who design recording equipment, can prove that these is no difference in tube or transistor sound. They do this by showing the latest specification sheets and quoting electronic figures which are visually quite impressive. It is true, according to the parameters being measured, that these is only a marginal difference in the signal quality. But are there some important parameters which are not being measured? One engineer who admits that there might be some marginal difference in the sound, says, "You just have to get used to the nice clean sound of transistors. What you've been listening to on tubes is a lot of distortion." Of course the question which comes to mind is, What is this distortion and how is it measured?

Psychoacoustically, musicians make more objective subjects than engineers. While their terms may not be expressed in standard units, the musician's "by ear" measuring technique seems quite valid. Consider the possibility that the ear's response may be quite different than an oscilloscope's.

"Tube records have more bass. . . . The bass actually sounds an octave lower," says one rock guitarist. A couple of professional studio players have pointed out on numerous occasions that the middle range of tube recordings is very clear, each instrument has presence, even at very low playback levels. Transistor recordings tend to emphasize the sibilants and cymbals, especially at low levels. "Transistor recordings are very clean but they lack the 'air' of a good tube recording." "With tubes there is a space between the instruments even when they play loud . . . transistors make a lot of buzzing." Two people commented that transistors added a lot of musically unrelated harmonics or white noise, especially on attack transients. This same phenomenon was expressed by another person as a "shattered glass" sound that restricted the dynamics. It was generally agreed that tubes did not have this problem because they overloaded gently. Finally, according to one record producer, "Transistor records sound restricted like they're under a blanket. Tube records jump out of the speaker at you. . . . Transistors have highs and lows but there is no punch to the sound."

When we heard an unusually loud and clear popular-music studio recording, we tried to trace its origin. In almost every case we found that the recording console had vacuum- tube preamplifiers. We are specific in mentioning preamplifiers because in many cases we found hybrid systems. Typically this is a three- or four-track console that is modified with solid-state line amplifiers to feed a solid-state eight- or sixteen-track tape machine.

Our extensive checking has indicated only two areas where vacuum-tube circuitry makes a definite audible difference in the sound quality: microphone preamplifiers and power amplifiers driving speakers or disc cutters. Both are applications where there is a mechanical-electrical interface.

(NOTE: Emphasis added. The above statement is one of the 2 most definitive in the study, directly answering the question posed in the title. Here Hamm essentially states that a definite audible difference from solid state is created when the tube is driving a speaker. The other major definitive statement is at the end of the study in part 3. BKB)

As the preliminary basis for our further investigation we decided to look into microphones and preamplifier signal levels under actual studio operating conditions. Hoping to find some clues here we would then try to carry this work further and relate electrical operating conditions to acoustically subjective sound colorations. Our search through published literature showed that little work has been undertaken in this area. Most microphone manufacturers publish extensive data on output levels under standard test conditions [1], but this is rather hard to convert to terms of microphone distances and playing volumes. Preamplifier circuit design is well covered for noise considerations [2], but not from the standpoint of actual microphone operating levels. Distortion has been treated in numerous ways [3-5], but with very few references to musical sound quality [10].

MICROPHONE OUTPUT LEVELS

To get a rough idea of the voltage output from different types of microphones, an oscilloscope was paralleled across inputs of a console. During the normal popular-music type sessions, peak readings of 1 volt or more were common, especially from closeup microphones on voice and drums. Due to the linear voltage scale, oscilloscope measrements over more than a 10-dB range are difficult. By building a simple bipolar logarithmic amplifier, the useful measuring range was extended to about four decades (Fig. 1). Considerable studio observation finally led to the construction of a peak holding type decibel meter. This circuit retained transient peaks of more than 50 microseconds within 2-dB accuracy for about 10 seconds; long enough to write them down. Using the logarithmic oscillo- scope display and the peak meter together proved very useful in gathering a wealth of data about real-life microphone signals.

Simplified bipolar logarithmic
amplifier schematic.

Fig. 1. Simplified bipolar logarithmic amplifier schematic.

Table I shows the normal peak outputs from several popular types of studio microphones. All the readings are taken with the microphone operating into the primary of an unloaded transformer. Pickup distances are indicated for each instru- ment and were determined by normal studio practice. Table II is an abridgement of a similar studio done by Fine Recording, Inc., several years ago. Details of this test setup are not available but the readings are probably taken without the 6-dB pad commonly used on the U-47 microphone today. Some calculations based on the manufacturer's published sensitivity for these microphones indicated that acoustic sound-pressure levels in excess of 130 dB are common. While the latest console preamplifiers have less noise, less distortion, and more knobs than ever before, they are not designed to handle this kind of input level. In most commercially available preamplifiers, head room runs on the order of +20 dBm, 1 and gain is commonly set at 40 dB. With these basic parameters it is clear from the data shown in Tables I and II that severe overloads can occur on peaks from almost all instruments. For example, a U-87 microphone gives a peak output of -1 dBm from a large floor tom. Amplification by 40 dB in the microphone preamplifier results in an output swing of +39 dBm, or almost 20 dB above the overload point. Logically a peak of this magnitude should be severely distorted.


Table 1. Peak microphone output levels for percussive sounds. Microphone Voltage, Open Circuit, dB Ref. 0.775 V

Instrument Distance (in.) U-87 U-47 77-DX C-28 666
Bass drum (single head) 6 0 -6 -9 -15 -1
Large tom tom 12 -1 -6 -9 -10 -5
Small tom tom 12 -1 -5 -7 -9 -1
Piano (single note) 6 -25 -29 -38 -35 -32
Piano (chord) 6 -23 -27 -36 -33 -33
Orchestra bells 18 -16 -25 -33 -33 -30
Cow bell 12 -10 -12 -29 -19 -15
Loud yell 4 0 -11 --- -10 -10

* U-87 and U-47 by Neumann, 77DX by RCA, C-28 by AKG, 666 by Electro-Voice.


Most recording consoles today have variable resistive pads on the microphone inputs to attenuate signal levels which are beyond the capabilities of the preamplifier. The common use of these input pads supposedly came about with the advent of loud rock music; however, this is not true in fact. For some 20 years it has been common to use a Neumann U-47 microphone for close microphone recording of brass and voice. Table II shows output levels requiring 10-20 dB of padding under these conditions, and this does agree with recording practice today where solid-state amplifiers are used. But most tube consoles did not have input pads and yet the same microphone performed with little noticeable distortion. Certainly brass players and singers are not that much louder today than they were yesterday. The microphone distance is about the same. The preamplifier specifications have not changed that much. Yet transistors require pads and tubes do not.


Table 2. Peak output for a U-47 microphone for various sounds.

Instrument Distance
(feet)
Peak
Pressure
Frequency
(Hz)
Microphone
Voltage (db
Ref. 0.775 V)
75-piece orchestra 15 350 -10
15-piece orchestra 10 350 -12
Trumpet 3 600 -16
Trombone 3 600 -15
French horn 3 300 -13
Flute 3.5 800 -26
Piccolo 3.5 2500 -18
Clarinet 3.5 350 -22
Bass sax 3.5 350 -8
Bass viol 5 150 -13


Here then is the hypothesis for further investigation. In the usual evaluation of audio preamplifiers it is assumed that they are operated in their linear range, i.e., harmonic distortion less than 10%. In this range tubes and transistors do have very similar performance characteristics. But the preceding section points out that amplifiers are often operated far out of their linear range at signal levels which would cause severe distortion. Under these conditons, tubes and transistors appear to behave quite differently from a sound viewpoint.

DISTORTION CHARACTERISTICS OF PREAMPLIFIERS

Three commercially available microphone preamplifiers of different designs were set up in the recording studio. Each amplifier was adjusted for a gain of 40 dB and an overload point of 3% total harmonic distortion (THD) at +18 dBm. Preamplifier 1 was a transistor design, preamplifier 2 was a hybrid operational amplifier, and preamplifier 3 was a vacuum-tube triode design. The amplifier outputs were terminated in 600-ohm loads and bridged by the monitoring system. The test signal, U-87 microphone, and large floor tom were switchable to each preamplifier input.

An informal group of studio personnel listened to the outputs of the three amplifiers on the normal control room monitor speakers. As the test signal was switched from one amplifier to another, listeners were asked to judge the sound quality. The output of amplifiers 1 and 2 was unanimously judged to be severely distorted. Amplifier 3, however, sounded clean. The test was repeated several times inserting attenuating pads in the microphone line until each amplifier sounded undistorted. Amplifier 1 could stand overloads of 5-10 dB without noticeable distortion. Amplifier 2 showed noticeable distortion at about 5 dB overload. Further listening revealed that it was only in the range of early overload where the amplifiers differed appreciably in sound quality. Once the amplifiers were well into the distortion region, they all sounded alike -- distorted. In their normal nonoverload range all three amplifiers sounded very clean.

The listening tests clearly indicate that the overload margin varies widely between different types of amplifiers. Engineering studios show that any amplifier adds distortion as soon as the overload point is reached. The tests show that all amplifiers could be overloaded to a certain degree without this distortion becoming noticeable. It may be concluded that these inaudible harmonics in the early overload condition might very well be causing the difference in sound coloration between tubes and transistors.

To get a general representation of the character of harmonic distortion in audio amplifiers, overload curves were plotted for about fifty different circuits. The tube circuits used the popular 12AY7 and 12AX7 triodes, the 8628 and 7586 triode nuvistors, and the 5879 pentode. These tubes have all been extensively used in recording console preamplifiers. The 2N3391A, 2N5089, and 2N3117 silicon PNP transistors were also chosen because of their extensive use in console and tape recorder circuitry. For comparison purposes tests were also run on the 2N5087 which is the PNP sister of the 2N5089. Operational amplifiers included the popular 709 and LM301 monolithic units and two commercially available hybrid designs used in recording consoles.


(Click here to continue reading this article)
Back to "The Big Idea"