Telephone Speech Transmission Impairments
Transmission media restricts the quality of telephone voice signals over analog lines. The longer the link, the worse the voice quality due to electrical impairments. This section examines the seven key factors that influence the quality of the received speech. The time period of focus is 1910-1965 when analog transmission was the backbone of long-distance telephone networks. Ten examples of audio impairments are included.
The diagram is a guide to the coverage of the speech channel deteriorations discussed below.
Introduction
The first official transcontinental telephone call on January 25, 1915, needed 2,500 tons of copper wire, 130,000 poles, and only three vacuum tube amplifiers in Pittsburgh, Omaha, and Salt Lake City. The link stretched nearly 3,400 miles from New York to San Francisco (333 Grant Ave.) [ATT], [Toll].
One can only imagine the received speech quality after such a journey. The analog impairments would be (1) signal loss, (2) noise, (3) distortion, (4) frequency response, (5) crosstalk, (6) echo and (7) delay.
​
The longer the trunk, the more necessary are the amplifiers and the need to keep the impairments in check. Some routes needed up to nine amplifiers to maintain the correct levels end-to-end.
​
The sections below illuminates each impairment using audio examples. Each instance uses a short audio clip mostly of Paul Harvey’s voice, a famous radio broadcaster and storyteller from the mid 1940's until 2008. His voice is used to simulate someone talking during a toll call. Here is Harvey's "reference voice", somewhat frequency limited to simulate the effect of a telephone call. Plus, it was recorded circa 1960's.
-
Repeating Amplifiers
All analog trunk lines of any design exhibit signal loss over distance. Loss is an irritation for every transmission engineer. So, amplifiers, often called repeaters, are needed for many toll calls.
Vacuum tube triode repeaters, invented at Bell Telephone after its purchase of the Audion rights from Lee DeForest, allowed telephone calls to travel beyond the unamplified limit of about 800 miles [Triode].
​
The transcontinental line required an amplifier every 1,100 miles on average. The reference [Toll] states a repeater is required every 100-150 miles for 4-wire trunks. The L-1 analog carrier system needed repeating every 8 miles (600 voice channels).
​
Bottom line, signal repeating is needed on trunks and the distance between amplifiers greatly depends on the trunk type, channel capacity and wire type. Without repeating the signal, the sound quality will fall below acceptable limits.
​
Listen here to an audio clip with sound level about 10% of the reference example above. This is how the volume as might sound on a long-distance trunk with a faulty repeater.
Mechanical audio amplifiers
​
Before the vacuum tube was sufficiently mature (about 1912) to be used in a production amplifier, mechanical amplifiers were relatively common for telephone trunk repeaters. What? How can a mechanical device amplify an electrical signal?
​
From [Fagen], “In 1903, H. E. Shreeve (Bell) was assigned the problem of developing a mechanical audio amplifier, using the receiver-transmitter principle. It was recognized by Shreeve that the requirements for such a device would be very severe, since it would have to be sensitive to small inputs, produce an adequate output, faithfully reproduce the input wave, and, finally, remain stable over long periods of operation.”
​
Shreeve’s audio amplifier is ingenious and seen in Fig 1. It uses the concept of a receiver portion stimulating a carbon granule transmitter portion, in series with a battery. The input signal modulates the armature, and this modulates the carbon to produce an amplified output signal. Very clever!
Fig 1, Schematic of basic mechanical amplifier after Shreeve. [Fagen]
Fig 2, Shreeve mechanical amplifier (circa 1904) and model 1A repeater (right) using the method. [Fagen]
The 3A (1912) repeater (improved 1A) represented the ultimate in mechanical amplifiers and was standard for Bell System use until the vacuum tube amplifier was introduced. Even the best mechanical amplifiers were limited to about three in tandem. So, their application in the telephone plant never became widespread.
Nonetheless, trunks need amplification and often required several in series. Even with the perfect amplifier, there were many challenges.
A Good Amplifier Puzzle
If you like a good puzzle, try this one.
A trunk line repeater must provide gain in each direction. Let’s call the directions westbound and eastbound. Amplifiers are one-way devices. So, is there a way using two amps to make this work? Fig 3 shows one possible way using two amplifiers in a back-to-back configuration. They are configured in a loop so an outrageous “singing screech” is all that will be heard.
Fig 3, How not to amplify a call in both directions simultaneously
Bell engineer Dr. George Campbell studied this problem in 1912, relying on mechanical amplifiers. He may have been influenced by T. Edison's work from 1879 on the Type 21 repeater. After experimenting with and abandoning the Type 21 repeater concept, Campbell invented the Type 22 “2-wire” repeater, Fig 4. A Type 21 indicates 2-way amplifying with 1 device while Type 22 indicates 2-way amplifying with 2 devices. The Type 21 is a very clever idea but difficult to implement. See Appendix A of the Toll Switching article for a short coverage of Edison's contribution to mechanical audio amplifiers and 2-way audio repeaters.
As an aside, Dr. Campbell became a famous contributor at Bell Labs. His work on inserting “loading coils” at regular distances on transmission lines became essential for improving toll speech quality over longer distances.
Fig 4, Schematic of Type 22 telephone repeater circuit
Each talking pair was connected to the input of one amplifier (mechanical or vacuum tubes) and the output of another by means of a so-called "hybrid coil," a transformer arrangement with four points (8 wire connections) of access. When the line was balanced by an identical artificial line, networks E or W, there was equal division of the signal between the real line and the E or W networks. Hence, no transfer of energy across the hybrid to the input of the other amplifier. -- Rephrased from [Fagen].
This neat trick prevented the amplifiers from feeding back on each other as they do in Fig 3. The singing problem was solved, in principle. Occasionally, there was a need to “re-tune” the E/W networks in Fig 4 on one or both sides to eliminate hints of singing and possibly echo. Unbalanced repeaters could become a frequent issue depending on how stable the East and West trunks were over time and temperature.
Singing repeater with no speech (simulated)
The 4-Wire solution
In 1912 Campbell proposed another radical idea. The amplifier portions of Fig 4 could be decoupled from the hybrid coils. This would create two, two wire paths, each with its own chain of simple amplifiers. There are many advantages to this approach compared to replicating Fig A4 for each 2-wire repeater station. See Fig 5 from [Duffy]. The ‘H’ in the figure is the hybrid transformer in Fig 4.
This idea was tested with mechanical amplifiers. It was applied in 1915 with electron tube amplifiers on a 450-mile cable circuit between Boston and Washington.
The test was very successful and helped kick off the building of the nationwide trunking network based on the 4-wire idea. The no. 4 Crossbar Tandem office was designed primarily for 4-wire trunks.
Fig 5, Converting 2-wire circuits to/from 4-wire circuits with amplifiers allocated in each leg of the trunk [Duffy].
The 4-wire idea had the drawback of requiring 2 pairs of wires for each trunk circuit. This was a costly tradeoff, and so only some trunks could warrant it for quality reasons.
Figure 6 shows a transcontinental line repeater (1915) with its door opened. It was located in a Pittsburgh suburb. The vacuum tube was on the front side. It’s likely 2 or more repeaters were on series at this station. The coast-to-coast route used 2 wires, not 4, for cost reasons. There were three repeater stations total on the 1915 line. It’s likely the hybrid coils are some of the round black cases.
Fig 6, Transcontinental line vacuum tube repeater at Brushton, near Pittsburg [Fagen]
Naturally, repeaters were needed worldwide. Fig 7 shows a two-wire telephone repeater, made by Standard Telephones and Cables (Britan) in collaboration with Western Electric Company, 1923-1925. Notice the two vacuum tube amplifiers (far right), one for each direction.
Fig 7, 2-wire repeater, London Science Museum
Repeaters kept improving and Bell, with others, introduced many varieties including some with automatic balancing compensation to keep singing under control.
2. Noise
Noise can come from many sources. To name a few -- induced from switching equipment (impulse noise), interference from nearby AC electrical devices/wiring, and repeaters (amplifier induced noise). In a talking circuit, instability of resistances (contacts, connectors) of less than one ohm may introduce objectionable noise. So, cable splices and terminal connections could create noise. Crosstalk is a form of noise but covered separately below.
​
Removing noise is nearly impossible so the strategy was to reduce it at the source. This demanded strict adherence to Bell standards of transmission quality. One important aspect was to use precious metal relay contacts (gold and the six so-called platinum metals) that do not readily oxidize or tarnish and consequently do not develop meaningful resistance variation. Silver was never used in the talking path (non-talking path, okay) due to its tendency to tarnish [Swenson] and cause noise.
Listen here to a call with several kinds of exaggerated induced noises.
As the voice path passes through an exchange, it swims in a sea of electromagnetic energy. With 60,000 relays in some central offices (per office-code) and large currents needed for some switching devices, it’s understandable that some of this radiated energy would find its way onto the talking circuits.
​
The older the exchange (Panel, SxS, #1 Crossbar), the more likely induced noises would be present while subscribers are making a call. This author fondly remembers the mysterious sounds heard after dialing the third and final digits on his panel office (552) in San Francisco.
​
The no. 1 Crossbar exchange (1938 introduced) generated an array of noises during call establishment. Listen below for the sounds immediately after the last digit is dialed. The "gonging" noises occurred when transmitting the number to the next exchange. Electromagnetic energy is coupled onto the talking path wiring from the relay switching actions. The caller would hear these sounds as the call progressed. There are six sequential calls in the recording. Compiled by ElmerCat.
3. Non-Linear Distortion
​
Non-linear distortion has output signal components not present on the input. New frequencies (harmonics for example) are generated. A distorted voice can sound fuzzy, garbled, raspy, and is mostly objectionable to the listener. By comparison, linear distortion modes do not add new spectral components. This type has spectral changes in amplitude or phase with no new frequencies added. It is generally less important as an impairment.
​
Fig 8 (left side) shows the spectrum of a single 440 Hz tone, with amplitude on the vertical axis. There is some broadening of the lobe (it is really a single vertical line at 440 Hz) due to the finite length of the test tone. This tone is then processed by a non-linear cubic curve to cause a distorted output. So, new “tones” appear (right side) at the 3rd (440*3 = 1,320 Hz), 5th and higher harmonics. This is what harmonic distortion looks like in the frequency domain.
Fig 8, Frequency spectrum of a pure 440 Hz tone (left) and a distorted version of the pure tone (right)
Listen below for an example of a non-linearly distorted voice.
What causes distortion? There are several origins. For example; signal clipping, “coil loaded” transmission lines [Fagen, pg 268], faulty repeaters, faulty (de)modulators/filters on analog carrier systems and nearly any input-to-output non linear transformation.
With the tens of thousands of devices, all part of the domestic toll network, it took high levels of quality monitoring and maintenance to meet Bell’s quality metrics.
​
Below is a voice distorted by a faulty demodulator (extreme case) in an analog multi-channel carrier system (L-1 Carrier, for example).
4. Frequency response anomalies
In general, there is a correlation between the money spent on transmitting voice signals and how much frequency bandwidth they occupy. Voice signals that are good enough for phone conversations are restricted to the range of frequencies with most of the speech energy -- from 200 Hz to 3.5 kHz. It’s not Hi-Fi, but good enough for every day conversations.
All voice trunks have some design limited frequency response. Ideally, every voice channel has “flat response” over the desired voice frequencies as seen in Fig 9 (left). The right image has linear distortion because the signal amplitudes vary compared to the desired flat response. The signal has effectively been filtered. This is very different from the non-linear “additive” distortion discussed above.
Fig 9, Amplitude/frequency responses ideal (left), degraded (right) [Engineering]
The resistance, inductance and capacitance inherently part of any transmission line filters the signal by changing the amplitude and phase of the signal components. It is unavoidable. Repeaters can restore what was lost by inversing the line filtering (equalization) and thereby partially restore what was lost.
​
Here is a sample of a voice that has been severely filtered by a long transmission line.
Finally, there is a type of frequency distortion called Delay Distortion. This is a linear-type distortion since no new frequencies are generated. For a 3,600-mile line (loaded coil, H-44-25 circuit) the difference in the delay at 1,000 cycles and at 3,000 cycles amounts to about 0.025 second. So, in effect, the frequencies of a speech wave will arrive at the listener’s ear at different times. Bell standards in the 1930’s limited this to 30 ms end to end [Clark] before speech quality starts to suffer.
The example to follow demos a "sweeping phase shift" (amplitudes and frequency positions are not affected, just phases) across the entire frequency band. This is a form of linear distortion.
5.Crosstalk
Crosstalk is a very common problem in shared infrastructure telephone systems. With possibly hundreds of calls simultaneously in the same office it’s inevitable that some of the conversations and switching noises will leak onto other talking paths. Adding to the woes is crosstalk from adjacent long-distance lines.
What are the mechanisms? One is “magnetic field induction.” This is a magnetic coupling. Another is “electric field induction.” This is an electric coupling. Together, these mechanisms induce unwanted sounds onto talking paths. Another means is a faulty carrier system (ex, analog L-1 type) that leaks the energy of adjacent channels or by some other mechanism.
During the 1920's-1950's mitigation strategies include spacing of the wires on polls, poll spacing, and the technique of transposing of talking pair wires at each poll. Proactive monitoring and removing under-performing trunks from circulation was a useful strategy [Chapman 1934]. Modern analog methods include twisted pairs, shielding of wires and equipment and ample grounding.
Here is an egregious example of crosstalk induced onto a conversation.
6. Echo
Echo is a nymph in Greek mythology. Hera, Zeus's wife, scorned her and she was sentenced to only repeat the last words spoken to her. Echo is a tragic figure with a cruel punishment. Nonetheless, it’s fitting to use the word echo to describe real-world sound reflections.
Many of us have heard the echo of our voice in a canyon, large room, or a tunnel. It’s fun to time the echo delay and then estimate the distance to the return wall. It’s not fun to hear our voice echoing during a telephone call.​What causes this unwanted event during a call?
It is well known that an impedance mismatch (a type of electrical discontinuity) along a transmission line or at the end point will produce a reflection back to the signal source. For short distances, less than ~55 ms of delay, listening is not impaired. For long distances, the delay can approach several hundred milliseconds and conversation is impossible.
For example, using a 1,000-mile circuit with a velocity of transmission of 10,000 miles per second (a coil loaded line, slows velocity << speed of light), the transmission time in one direction is .1 second and for a round trip it’s .2 second. See [Clark and Mathes] for many insights on early echo issues and related suppressors. This reference is the gold standard for the early echo work.
When the sent signal arrives at the end point, some of its energy may be reflected based on how well the signal is "absorbed" by the end point network. The poorer the absorption the more an echo is generated. The longer the end-to-end delay, the more the echo needs to be attenuated. The metric for this is “echo return loss”. So, for a 50 ms delay the return echo attenuation should be at least 40 db. For a 500 ms delay, the value is closer to 60 db.
Listen below to a call with a glaring example of a 500 ms round trip delay, in one direction only. If this was your conversation, you would slow down considerably or hang up and try again. The level of the return echo and the roundtrip delay are the most important metrics. In this example the male speaker does not back off during his echo although a slow down would normally occur. Of course, the echo could be in both directions. With an echo suppressor inserted, discussed below, there would be only a small or no echo if the system was aligned correctly.
Echo suppressors
Since some echo is all but inevitable on raw long distance wire trunks, mitigation techniques are required. Echo suppressors go a long way toward eliminating echo effects but did not remove them entirely especially during the 1900-1975 period. For the most part, echo remains the most important limiting factor on the long-haul lines after attenuation.
Fig 10, Basic operation of an echo suppressor in a 4-wire circuit (redrawn by author)
Figure 10 [Fagen] shows a simplified Echo Suppressor consisting of two relays connected such that the return path of the 4-wire circuit is short-circuited whenever speech is detected in the forward path. In the figure, West is talking, the “W-E speech detector” causes its relay to engage thereby shorting out the echo from the East station. Parts of the Fig 10 circuit are related to 2-way, 4-wire, voice repeaters. See Figs 4 and 5 above.
Occasionally when the speakers at both ends utter words at nearly the same instant both echo suppressors respond. Each suppressor’s relay blocks one direction of the circuit so no side can hear. So, suppressors are placed at least every 100 ms of delay else this problem is magnified, and call quality suffers.
A number of Bell System designed echo suppressors were first placed on 4-wire lines in Harrisburg, Pennsylvania, during 1924. See Fig 11. Techniques were also developed to echo suppress on 2-wire lines. The first production model was the Western Electric 1A Echo Suppressor.
Surprisingly, this brute force method of alternately shunting the echoes with a relay or equivalent was used until the 1970’s. John R. Kelly (Bell Labs) invented a completely new approach and was granted US patent 3,500,000, Self Adaptive Echo Canceller. This method uses the east(west) bound speech and compares it to the west(east) reflected echo and subtracts one from the other. Cancellation was perfected when the phasing, amplitude and delay of eastbound and westbound signals were accounted for. It proved a grand idea and was the future of echo cancelling technology. See [Duffy] and [Brady].
Fig 11, 4-wire echo suppressor rack of 8 units, 1924
7. Delay
When delay is introduced into a two-way telephone conversation between people who are inexperienced with delayed transmission, measurable changes occur in their speaking behavior. The subjects become confused more often, engage in more double talking and mutual silence, and exhibit certain changes in their on-off pattern generation behavior. For a round-trip delay of 600 ms, the above effects reach their asymptotic values. The measurable quality of a call degrades rapidly where the mouth-to-ear delay latency exceeds 200 milliseconds. Of course, echo and delay are related but delay alone is problematic. Rephrased from [Brady].
​
Conclusion
​
Keeping telephone voice impairments in check was crucial for clear and effective transmission. The seven factors cited above each degrades the quality of a call. One or more of the impairments often occurred simultaneously. Regular network monitoring ensured that the transmission quality met or exceeded Bell's standards.
In North America, trunks were all analog until about 1962. The introduction of the T-1 (T-2, T-3) digital carrier system started the transition to digital. All the impairments discussed in this article affect digital less than analog transmission methods. So, it was easier with digital to maintain high speech quality and analog trunks gradually became obsolete.
References
​
Brady, P. T., Effects of Transmission Delay on Conversational Behavior on Echo-Free Telephone Circuits, Bell System Technical Journal , January 1971
Chapman, A.G., Open-Wire Crosstalk, Bell System Technical Journal, April-June, 1934
Clark, A.B., H.S. Osborne, Long Distance Telephone Circuits in Cable, Bell System Technical Journal, Oct, 1932.
Clark and Mathes, Echo Suppressors for Long Telephone Circuits, Journal of the A.I.E.E, June 1925. Both authors are from Bell Labs.
​
Duffy, F,P. et al, Echo Performance of Toll Telephone Connections in the United States, Bell Laboratories Record, Feb 1975.
Engineering and Operations in the Bell System, AT&T Bell Laboratories, Ed 2, 1982
Fagen, M.D., A History of Engineering and Science in the Bell System (1875-1925), Early Years, 1975
Swenson, P.W,, Contacts, Bell Laboratories Record, Feb 1949
Toll: Technical Developments Underlying the Toll Services of the Bell System, Bell System Technical Journal, Oct-Dec, 1936.