Up one level Back to Home

An Explanation of "Flat Audio", "Pre-Emphasis" and "De-Emphasis" By Mike Morris WA6ILQ, Jeff DePolo WN3A, Kevin Custer W3KKC and Bob Schmid WA9FBO Created from an email discussion in 1999

As background information, you may wish to look at this page on FM theory first.

Comments on the topics brought up in this discussion are quite welcome.   This web page currently has four authors.
If you have a contribution to make to this page, please send it in. to Mike WA6ILQ at (callsign) /at/ repeater-builder /dot/ com.
We have no problem with the idea of expanding it to additional authors - even five or even six authors.

An Explanation of "Flat Audio" in the
NBFM Repeater and Link Environment
By Michael R. Morris WA6ILQ
Originally written January 1999, and updated several times since

The early days of FM radio were all broadcast related and all the transmitting equipment used what we amateur radio operators would consider very wide modulation (75 kHz) and "phase modulation" (PM) circuitry to produce the FM signal... the modulation circuitry was simple and worked well.   One inherent characteristic of the phase modulator was (and still is) that the deviation doubled with the audio frequency.   In other words, if you set the modulation level for 5 kHz deviation at an audio frequency of 1 kHz and then doubled the audio frequency to 2 kHz you'd find that the deviation was doubled as well to 10 kHz.   At 4 kHz the deviation would be doubled again to 20 kHz.   Each doubling is an increase of 6 dB. This inherent 6dB doubling was already well-known in the broadcast engineering world and was called "Pre-Emphasis" in the transmitter and required a compensating "De-Emphasis" circuit in the receiver.   There was no way around it - the pre-emphasis characteristic was an unavoidable part of the transmitter modulator and you had to have a matching de-emphasis circuit in the receiver.   A true frequency modulator that was easily mass produced and didn't require frequent adjustment (i.e. one that was practical outside the laboratory) was not developed until much later (when varactor diodes became commonplace).   This pre-emphasized audio characteristic and 75 kHz deviation are two reasons that the broadcast FM channels are to this day spaced 200 kHz apart.   And the early developers of broadcast FM didn't try that hard to eliminate the phase modulator pre-emphasis as the de-emphasis circuitry in the receiver not only restored the transmitted audio back to normal, but in the process also reduced the high frequency background noise hiss in the receiver.

So now we move into the 2-way radio environment where narrow band FM is the rule, FM channels are spaced as close as 15 kHz, sometimes 12 and a half kHz and 6 and a quarter kHz spacing is coming.   Most crystal based radios used PM, and since varactor diodes are used in the VFO section of a synthesized radios you will find that most synthesized radios used FM with pre-emphasis stages in front of them for compatibility.   In fact, to this day you will find that conventional voice communications are all pre-emphasized and de-emphasized where most data and digital (i.e. D-Star, P25, etc.) communications are not. Early packet ran at 300 baud, and both it and 1200 baud packet is run as audio tones and uses pre-emphasis and de-emphasis, 9600 and 19.2k baud is generally run digitally and without pre-emphasis and de-emphasis. More notes on packet, APRS, pre-emphasis and de-emphasis is here (an off-site link).

In a conventional simplex (non-repeating) environment the transmitting user's transmitter pre-emphasizes the audio, and the receiving user's receiver de-emphasizes it, reproducing the original audio.   When the user switches to a repeater, the same thing has to happen, but now with four radios in line instead of two: the transmitting user's radio, the repeater receiver, the repeater transmitter, and the receiving user's radio.   Most repeater systems de-emphasize in the system receiver, process the normal audio in the repeater controller and then pre-emphasize in the system transmitter, which works just fine if the two emphasis curves are matched (most factory ones aren't, and the reasons why will be explained later). Other systems take advantage of a true FM modulator and run the audio unchanged through the repeater hardware - they take the receiver's IF frequency, demodulate it then take the raw audio and then inject it into the transmitter modulator and the only processing is an audio gate circuit to mute the squelch noise, an audio limiter to avoid over-deviation, all followed by a low-pass filter to get rid of the artifacts created by the limiter plus the high frequency components of the squelch noise burst (the squelch tail).   In a system running CTCSS (i.e. PL, or CG) there might a high pass filter with a knee at about 200 to 250 Hz to strip off the user's sub-audible tone.   Note that systems that do this still appear to be a "normal" repeater since at the RF level it takes in pre-emphasized communications and outputs pre-emphasized communications.

My personal choice is to run flat audio inside the repeater cabinet as much as possible so that the transmitting users radio does the pre-emphasis of the audio, the repeater receiver leaves it alone, the repeater controller leaves it alone, the repeater transmitter leaves it alone and the receiving users radio does the de-emphasizes. Any distortion or other issues with the audio is the fault of the user radio(s).

The frequently misunderstood term "Flat Audio" refers to the fact that the audio path through the repeater is spectrally flat, neither accentuating the highs or the lows.   There is no pre-emphasis or de-emphasis in the audio chain between the repeater receiver discriminator (demodulator) and the repeater transmitter modulator.   The confusion surrounding the term "flat audio" comes from the fact that the audio path, while flat, carries pre-emphasized audio - this does not change the fact that the path itself does not de-emphasize and re-pre-emphasize.   The path may be flat but the audio it carries is NOT, and won't be until it is run through a de-emphasis network somewhere (most commonly the end-user's receiver).   As a result the "flat" repeat audio path in the controller does, however, require that any audio "destinations" in the controller, like the touchtone decoder, the receiver side of the autopatch, etc. need to have de-emphasis components added.   Also all local audio sources like the local microphone, the controller speech synthesizer chip, the controller ID'er and status tone generators, the transmitter side of the autopatch, etc. have to be properly pre-emphasized before they hit the audio mixer that feeds the repeater transmitter, otherwise these sources will not sound natural after the final de-emphasis in the user's receiver.

Another way of stating the above is that the process of taking audio out of the receiver, right at the detector / discriminator where it is still pure and pre-emphasized by the originating user's transmitter, passing it through the audio path of a properly designed and configured repeater controller, and then inject it into the transmitter well past the microphone input, past the speech amplifier, past the pre-emphasis network, and directly into the modulator keeps the audio "flat" and undistorted through the entire repeater system (the repeater receiver local speaker and repeater transmitter local mic are unmodified).   In a flat audio repeater, the relative levels of a 1 kHz and a 2 kHz tone will be exactly the same throughout the audio chain (from the discriminator to the modulator).   Note that I say RELATIVE levels.   They may go up or down, but they will both go up or down by the exact same proportion.

Note that improperly done "flat audio" mods will bypass the necessary filtering and limiting in the transmitter (the flat audio injection point is after the filtering and limiting internal to the transmitter chassis).   Most transmitters regardless of manufacturer will splatter all over without the proper filtering and limiting.   You can easily trash the adjacent channel (or more) on each side of your transmitter.
The goal is to take flat audio from the receivers and pass (with less than 0.5 dB ripple across the band from 275 Hz to 3100 Hz.

FCC Rules Part 97.307(b) says "Emissions outside the necessary bandwidth must not cause splatter or key-click interference to operations on adjacent frequencies."

Some proponents of "flat audio" say that they do it because they want to make their repeater system "sound like simplex".   Well, yes, but that may be a misleading comparison to some people: have you listened to simplex recently?   The average fresh-out-of-the-box Japanese radio is NOT set up properly, many are over-deviated, and people either whisper from two feet away or they hold the microphone against their lips and shout into it.   Audio processing in the system can help, but a better solution is to hold a club meeting and do what the production line should have done - you will need to have a tune-in / tune-up session with a properly calibrated service monitor operated by somone that knows what he is doing. And in the cases of those that shout in the microphone and / or get right up on top of it you may have to add a microphone level pot inside the radio or inside the microphone to lower the overall audio level to something reasonable.

Once your repeater has a user base with properly set up radios, you can think about adding a little audio processing to your repeater system.   At that time I suggest that you read the audio processing article written by Jeff dePolo WN3A, a broadcast engineer for a large group of FM stations.

By the way, the term "flat audio" is not reserved to the amateur community, they acquired it from the commercial land mobile industry.   In fact, you can order a "Flat Audio" card that plugs into the Motorola Spectra TAC receiver chassis.   The card is identical to the regular card except that it is missing the de-emphasis components.   The title of the manual section specifically says "... modified for use in "flat audio" situations".   Another example is that the MICOR base station had a card, the TRN5439B Flat Transmit Audio Board that adapted it for zero-pre-emphasis (normally used in paging transmitter environments).

Back to pre-emphasis and de-emphasis:
One point I'd like to stress (and yes, I'm repeating myself): the pre-emphasis stage is integral with the modulator in a phase modulated transmitter - you cannot separate them no matter how much you'd like to.   In a true frequency modulated transmitter the pre-emphasis is separate from the modulator and you want to leave the pre-emphasis in line for the local microphone, and inject repeat audio at the point between the pre-emphasis and the modulator.   Some of the phase modulated Motorola and GE transmitters can be modified for true FM.

The design of the (unmodified) two-way radios show that they are designed for mobile operation. The features include pre-emphasis, deviation limiting (clipping) and receiver de-emphasis, and "forgiving" squelch operation (due to the fact that the user radios (stations, mobiles and handhelds) will have flutter and widely varying signal strength).   If you use stock radios to implement the center point in a long distance point-to-point link (i.e. back-to-back radios) you will add distortion, since the mismatched paired de-emphasis / pre-emphasis at each hop point will add measurable distortion to the total end-to-end path as the length / site count increases.   For multiple-hop links, these stock radios can add gross problems, such as excessive distortion (due to mismatched pre- and de-emphasis), audio frequency response being very poor and very long squelch bursts (compensating for mobile flutter and fading).   All these conditions will cause a point-to-point linking system to operate very badly and be rather annoying and fatiguing to listen to.   In a point-to-point relay situation you can rework the squelch and audio circuits since you will be handling consistent strength signals with no mobile flutter (see this offsite article on modifying Mitreks for use as link radios for an excellent overview).   Unless extreme care is used and the de-emphasis and pre-emphasis circuitry is built with high quality close-tolerance components every audio stage will add additional distortion (most common resistors are +10% / -10% accuracy (and some manufacturers save a penny everywhere they can and instead use the cheapest parts they can get away with).   Where common resistors are + / - 10%, the common capacitors have a + / - 20% accuracy, and some are as poor as +100% / -60%, so identical circuits with "identical" production parts will rarely have identical performance characteristics).

The manufacturers don't try and hide their sloppy efforts or failures at matching the pre-emphasis and de-emphasis curves - even Motorola says in multiple sales brochures (example: in the Spectra Motorcycle brochure, available on this web site) and in many common mobile and station manuals (example: 6881040E80, for a Mitrek tabletop base station, also available at this web site) that the audio response curve is "+1, -3 dB of a 6 dB/octave pre-emphasis characteristic from 300 to 3000Hz".   In other words, they can have a circuit that is as poor as 50% of the designers goal and still meet their specification (because they made their specification overly loose).   As another example of sloppy work, a while back two friends of mine ended up with "sister" handhelds (sequential serial numbers) from a major Japanese manufacturer.   Not only was the out-of-the-box microphone gain and deviation totally mis-set (and the two radios were very different even to an untrained ear), but one transmitter was noticeably tinny compared to the other.   This is one reason why anyone who is setting up a repeater needs to beg, bribe, or somehow acquire the use of a service monitor for the initial setup.   You can't be certain that any random handheld or mobile (i.e. your own radio) is appropriate as a primary standard to set up the audio so the system "sounds good".   The pre-emphasis and de-emphasis curves in your radios may not be anywhere close to 6 dB per octave.   A service monitor is built with precision parts and is.

Another example of what not having proper test equipment can do: Years ago I was involved with fixing a system that was horribly tinny - but the owner kept insisting it "sounded just fine". It was his own hearing that was off. Once he got some good hearing aids with some proper audio compensation he could hear just how bad he had set up his repeater.

Another characteristic that adds system distortion is that some capacitors will change value with age, and usually will change value with temperature.   The pre-empasis/de-emphasis component selection has to be done with unheated / uncooled hilltop radio sites and extremely low temperatures in mind (northern Canada in winter) or similar high temperature extremes (an elevator equipment room on the top of a building in Phoenix or Dallas in the summer).   And don't forget the fact that the average ham is downright cheap - and have been known to mix and match whatever radios they can get in quantity to build their linked systems.   One site may have a Brand M radio for an eastbound link, a brand G radio for a westbound link, a Brand K for a local repeater and a brand J for a remote base.   All of the literature will claim to have the same 6 dB curve, but if you go looking you will find that it is implemented with different RC ratios, and if both the Cs and the Rs are 20% off, and both will drift with age, and with temperature (and most radio sites are not climate controlled).   Now stack several hops of mismatched de-emphasis / pre-emphasis end-to-end (due to age, temperature or loose tolerance parts) and while each radio sounds pretty good by itself all of the marginal tolerances add up and the total end-to-end audio will sound really bad.

It's not unusual for the system controllers in these large linked systems (usually designed and constructed by the groups themselves) to have audio performance equal to or exceeding premium quality hi-fi equipment - and implemented (and tested) over a wide temperature range (I'm referring to harmonic distortion, phase linearity, group delay, and similar parameters, but limited to the 200 Hz to 5 kHz communications audio spectrum).   Now you know why large linking systems with many hops will prefer flat audio (and minimal processing) in each controller.   They believe that the less audio processing that is present in each piece of equipment, the cleaner the overall system is...   and it usually is !   Point of information: According to the map that I saw the Cactus Intertie in the southwest USA has 25 hops between Los Angeles, California and Corpus Christi, Texas. The audio sounds as good as if it was local.

Sidebar:

As far as I know at the moment there are no commercial multi-port controllers that meet these requirements:
1) the ability to select (i.e. with jumpers) absence or presence of de-emphasis or pre-emphasis networks in the full duplex repeat audio / link audio path
2) precision de-emphasis or pre-emphasis networks (i.e. none of this +1 to -3 dB accuracy), and with stable long-term performance over a wide temperature range
3) a high performance squelch circuit per radio port
4) a 16-button touchtone decoder per radio port
5) the ability to select (i.e. with jumpers) absence or presence of CTCSS filters on a per port basis
6) an audio notch filter to remove link IDers
7) hi-fi-grade audio characteristics (and measurable and repeatable across a wide temperature swing)
8) addressable controllers (i.e. can select which controller you are accessing when you have many (even over 100) on line at once)
9) status telemetry that can report back all the configuration parameters over a link port.

If you do chose to keep the repeater's stock receiver de-emphasis and transmitter pre-emphasis in line, you have to keep them matched and paired (which may take replacing some aged parts - and using 1% parts as the replacements).   If the pre and de-emph circuits change the audio using the same time constant... there should be no real world changes to the original voice audio.   Having matched de-emphasis or pre-emphasis networks within the repeater's voice audio path should not result in any detectable difference in the audio path.   There is audio spectrum analysis software and sweep generator software that can be run on a computer that has a top quality sound card to test this - but first run a test of the generator and analyzer software with just a jumper cable / loopback cable between them.   I've seen a number of "premium" sound cards that honestly weren't.
And a pair of Motorola Astro-Spectra UHF radios were measured as 6 and 7db per octave pre-emphasis and 4 db per octave de-emphasis.

Another problem that rears its ugly head unless you know the equipment you are working on intimately... If you pick off raw (i.e. not de-emphasized) audio from the receiver discriminator and pipe it into the microphone jack of a transmitter you will end up with an extra level of pre-emphasis (commonly called "double pre-emphasis") that will cause the audio to sound very tinny or shrill (take your home hi-fi, tune to a talk radio station, center the bass and the treble controls, note the audio characteristics, then crank the bass control to minimum and the treble to maximum - and mentally double or triple the overall effect).   On a true FM transmitter you can sometimes bypass the pre-emphasis network, on a phase modulated transmitter there is no way around it without adding a de-emphasis network in front of it to compensate.   This is why many repeater controllers have a built in de-emphasis network that can be jumpered into the circuit or jumpered out as needed.   You will find a de-emphasis option on the receive side of the controller much more often that you will find a pre-emphasis option on the transmitter.   On those that do not have pre-emphasis you can add it with the AP-50 board from Repeater-Builder (the company).

Likewise, picking audio from the receiver after the de-emphasis network (in some receivers that point is after the volume control and the audio muting part of the squelch circuit) and piping it into a true FM transmitter modulator can produce audio with extra amount of de-emphasis (commonly called "double de-emphasis") resulting in a very muffled, bassy sound with no high frequencies (same example as above, but crank the bass control to maximum and the treble to minimum - and mentally double or triple the overall effect).

The mismatched de-emphasis / pre-emphasis situation is common in the packet world... people take the transmit audio generated by their TNC (think "modem") and inject it into their microphone jack (i.e. that audio gets pre-emphasized), and they feed recevier discriminator audio (i.e. non-de-emphasized flat audio) to the receive side of their TNC. And then they wonder why they have short range and lots of resends / retries and timeouts.
The fix? Use both de-emphasis and pre-emphasis, not just one. If you are using discriminator audio for receive then add the correct R-C combination between the discriminator and the TNC. For correct values please see this article.

Either of the above two situations (double-de-emphasis or double-pre-emphasis) is instantly recognizable by an experienced ear.   This situation gets especially interesting when you connect an HF remote base to a "flat audio" VHF, 220 MHz, UHF, 900 MHz or 1200 MHz repeater controller as HF does not use pre-emphasis / de-emphasis and everything that runs narrowband FM does....

One location where the "flat audio" method of coupling the receiver to the transmitter is frequently used is in the outlying receive sites that are part of a voted repeater system.   One method that is used to cut down on the number of sub-audible tone decode and encode delays is to run the satellite receiver site as a flat audio system even if nothing else in the system is. Take the main channel receiver discriminator audio and pipe it into the link transmitter modulator with minimal processing, and run the entire site in carrier squelch (it's a link channel, nobody is going to listen to it except the voting / link receivers).   Make sure that the audio curve is flat from just below whatever CTCSS tone you are using on up so that the end-users CTCSS encoder tone passes through with no distortion.   The proponents of flat audio claim that bypassing the unmatched pre-emphasis / de-emphasis components (the link radios are frequently of a different manufacturer than the main channel receivers) results in a "cleaner" link and better voting performance.   They may very well be right.   As a bonus this system design lowers the overall CTCSS pick-up delay by placing the first CTCSS tone decoder at the voting panel end of the satelite receiver point-to-point RF link.   With careful link receiver and voter configuration and design you can run one CTCSS tone decoder behind the voting panel...

Another point - if you have a digital recorder / playback system (sometimes called a Digital Voice recorder or DVR) in your repeater controller you need to be careful in the setting of the normal / de-emphasis jumpers on the record side.   If you have a repeater transmitter with pre-emphasis then you will need to feed normal audio to the recorder.   If you have a flat repeater transmitter then you will want to feed the record side with de-emphasized audio.   Many controllers have a normal / de-emphasized jumper in the receiver port area (the Arcom RC210 and Scom 7330 are two) and if you have the jumper set for de-emphasized and record a track, then play it back through a pre-emphasized repeater transmitter then the user is going to hear double-de-emphasized audio and it will sound really muffled and "muddy".

Two important concerns that I've totally ignored in the above writeup is audio processing and impedance matching.   Processing can be done in the transmitter stages between the microphone and the modulator, and similarly in the stages between the discriminator and the repeat audio output point (which may or may not be the same audio that is fed to the local monitor speaker).   Impedance matching has to be done between stages in the receiver, in the transmitter, and in the repeater audio panel / repeater controller that connects them.   Improperly handled, a screwup in any stage of either process can ruin the otherwise carefully managed audio.

And before I forget, DTMF, also known as TouchTone, is made up of two groups of tones.   A screwup in the matching of the pre-emphasis and de-emphasis curves can result in an objectionable amount of DTMF "twist".   The term "twist" in the telco / telecom community is used to quantify any unwanted difference between the levels of a lower frequency audio signal and a higher frequency audio signal, no matter if they are simultaneous (like DTMF) or not.

The DTMF keypad is laid out in a 4 row by 3 column 12-button matrix (or optionally 4 columns / 16 buttons). Many repeater system owners prefer the 16-button encoders and decoders as it offers much more flexibility in setting up the repeater controller commands.   Each row on the DTMF matrix is a low group tone frequency, and each column is a high group tone frequency.   Pressing any single key (such as "1") will send both tones simultaneously (697 and 1209 Hertz).   On most DTMF dials pressing two keys in any row or two keys in any column selects the single tone in that row or column (i.e. pressing any two of 1, 2, 3, or A will cause the dial to send the single tone 697 Hz).

	1209 Hz	1336 Hz	1477 Hz	1633 Hz
				Civilian Labeling	Military Labeling
697 Hz	1	2	3	A	Fo
770 Hz	4	5	6	B	F
852 Hz	7	8	9	C	I
941 Hz	* "Star"	0 Zero	# "Pound"	D	P

The four "row tones" are commonly called the "low frequency group" or "low group" and the four column tones are commonly called the "high frequency group" or the "high group".

An early reference book had a critical typographical error - the 1477 Hz. tone was listed as 1447 Hz., and that error has continued in some documentation.   Be assured that the third column is really 1477 Hz.

The "real" telephone company dial pads had built in  /  designed-in twist that accentuated the high group by a few dB, it was there on purpose and was designed to help overcome the capacitance of the telephone cabling between the subscriber set and the central office (it's amazing what the capacitance in several miles of 26 gauge twisted pair wire will do to audio).   Many of the DTMF encoder chips have the "telephone line" preset level of twist built in, the application notes for the Motorola 14410 DTMF encoder chip showed how to implement a trimpot-adjustable twist adjustment that could be set to zero for use on radio circuits. The data sheet shows a pair of 20K resistors used to create an equal-level audio mixer. Simply use a 50 K pot instead, with the two ends connected to the chip outputs and the wiper as the mixer output. Then adjust the pot for the level of "twist" that you want (zero).

The tone frequencies used in DTMF were well thought out, but not far enough.   Any distortion between the source and the decoder will result in intermodulation products being generated and the result will cause unreliable decoding.   For example, use the classic 2A-B intermodulation calculation: The DTMF star button is 941 and 1209 Hertz.   Plugging in the 1209 frequency for "A" and 941 as "B" results in 1209 times 2 = 2418, minus 941 = 1477.   The DTMF pound button is 941 and 1477.   So any distortion in a transmission path can result in a "star" being decoded as a "pound", not being decoded at all, or even as a "star" and a "pound" simultaneously.   This is one example why system designers have to keep transmission paths linear and distortion free.   Otherwise the system builders will blame the repeater controller manufacturer.   And when the controller goes back to the manufacturer for "repair" he finds nothing wrong with it.   Because there isn't anything wrong with the controller.

And it's not common knowledge that the DTMF tone chart is larger than four rows and four columns.   If you extend the tone mathematical sequence it works out to an 8 by 8 matrix.   Yes, the Bell System / Western Electric engineers designed it as a 64-button system / matrix and only the top left corner was ever implemented.   I've seen a xerox copy of some original TouchTone notes, the 8x8 math is very interesting.

From the description above you can see that a situation where the de-emphasis network does not match the pre-emphasis network can result in the levels of the two groups of tones being different, as the pre-emphasis raises the level of the high group to a large degree above the low group and it will be most noticeable on the "A" pair of tones.   If the amount of twist is greater than the twist tolerance level of the DTMF receiver the system will fail to decode the button presses - and most users and system operators will blame the repeater controller manufacturer, not the user radio and the repeater receiver.   And when the controller goes back to the manufacturer for "repair" he again finds nothing wrong.   A good field test for twist is to put an AC voltmeter (even your VOM set on a low AC volts scale) across the repeater receiver speaker leads.   Then press the PTT on your handheld and hold down the DTMF "1" and "2" keys simultaneously.   This sends the 697 Hz tone by itself.   Set the receiver speaker volume control and the VOM range switch so that the AC voltage reads mid-scale on the VOM.   Now switch to holding down the "A" and "B" keys to generate 1633 Hz by itself. The VOM reading should be very, very close if the pre-emphasis and de-emphasis match.
Please realize that this is a quickie, very minimal test - it's only at two audio frequencies, and assumes that:
a) the DTMF encoder audio level is the same on both of the two audio frequencies,
b) the audio response of the transmitter audio chain before the pre-emphasis stage is the same on the two frequencies,
c) the receiver audio post-de-emphasis response is the same on the two frequencies,
d) and that the AC Voltmeter is reading correctly on the two frequencies.
If the two levels are very close then it's worth trying all 8 DTMF tones - or even borrowing some real audio test gear. As I said above, an audio sweep generator and an audio spectrum analyzer (even one that is software driven and based on a computer sound card) can be a big help in evaluating receivers, transmitters, and modifications to them.

/s/ Mike Morris WA6ILQ   Originally written January 19 1999, revised several times since.

An Explanation of "Flat Audio"
By Jeff DePolo  WN3A

Flat Audio:
When most of us talk about "flat audio", we're talking about passing audio through the repeater without any de-emphasis/pre-emphasis stages in line.   Discriminator audio that has already been pre-emphasized in the user's radio gets passed to an FM (not PM) modulator without any intervening de-emphasis in the receiver or pre-emphasis in the exciter.   Of course there should still be limiting and low-pass filtering.   Most controllers can be adapted to handle "flat audio" (i.e. pre-emphasized audio from the discriminator).   This usually involves adding some de-emphasis on the audio going into the DTMF decoder and the send side of the autopatch interface, and some pre-emphasis on the speech, tones, and receive side of the autopatch interface.

So in reality, "flat audio" is somewhat of a misnomer, since the audio you're dealing with through the repeater isn't flat at all, it has been pre-emphasized in the user's radio.   My guess is the term caught on because people were bypassing the stock audio circuitry (de-emphasis/filtering in the rx and pre-emphasis in the tx) to make the overall audio response "flatter" in terms of minimizing the low-end and high-end rolloff.

Flat audio is one of those phrases that can be used two different ways that are polar opposites of each other.   In narrowband FM like we use on amateur, and commercial repeaters, the user's transmitter pre emphasizes the audio at a rate of 6 dB per octave.   In laymen's terms, the higher frequencies are transmitted at a higher deviation than the lower frequencies.   If you were to listen to "raw" FM audio that has been pre-emphasized, it would sound tinny.

In a normal repeater, the audio that was pre-emphasized in the user's radio gets de-emphasized in the repeater receiver, thus returning it back to normal status.   The audio then goes through the controller or other audio stages in the repeater control shelf.   When it gets to the repeater transmitter, it gets pre-emphasized again.   On the receiving end, the user's radio de-emphasizes the audio that was pre-emphasized by the repeater transmitter, returning it to normal audio on par with what went into the originating station's microphone.

However, the de-emphasis in the repeater receiver and the pre-emphasis in the repeater transmitter can be eliminated together as a pair since they are reciprocal.   What gets received by the repeater receiver as pre-emphasized audio can go back out the repeater transmitter with the original pre-emphasis kept intact.   This is what is often referred to when people talk about "flat audio mods" in a repeater installation.   In reality, whether there is a de-emphasis/pre-emphasis pair done in the repeater, or whether the repeater skips the de-emphasis/pre-emphasis steps, a repeater always repeats "flat" audio.   It comes in pre-emphasized and it goes back out pre-emphasized either way.

There are many considerations to be made when skipping the de-emphasis/pre-emphasis steps in a repeater, primarily those pertaining to audio sources aside from the repeater receiver (such as the controller's synthesized speech, the controller's tone generators, audio to/from the autopatch, etc.).   In such cases, these ancillary audio sources need to be pre-emphasized separately, something that most controllers don't provide provisions for doing, since the transmitter won't be providing the necessary pre-emphasis on its own.

There are other technical considerations with regard to how to get around the stock audio processing in repeaters as well as the pre-emphasis/de-emphasis stages in the receiver and transmitter.   These aspects require particular attention to detail, especially surgery to the transmitter which can easily result in causing interference in the form of adjacent channel splatter, spurs, and broadband noise.

Jeff DePolo WN3A, January 19 1999

An Explanation of "Pre-emphasis, & De-emphasis"
By Kevin K. Custer W3KKC

Emphasis:
The concept of Pre-emphasis and De-emphasis is a broad subject, however I will try to give you the basic concept behind the advantages, and necessities (or lack of them) in NBFM.

In common narrow band two way fm communications,  Pre-emphasis follows a 6 dB per octave rate. This means that as the frequency doubles, the amplitude increases 6 dB. This is usually done between 300 and 3000 cycles. Why is it necessary?  Pre-emphasis is needed in FM to maintain good signal to noise ratio.  Common voice characteristics emit low frequencies higher in amplitude than high frequencies. The limiter circuits that clip the voice to allow protection of over deviation are usually not frequency sensitive, and are fixed in level, so they will clip or limit the lows before the highs. This results in added distortion because of the lows overdriving the limiter. Pre-emphasis is used to shape the voice signals to create a more equal amplitude of lows and highs before their application to the limiter. The result is that the signal received is perceived louder due to more equal clipping or limiting of the signal, but probably more important is the increased level of the higher frequencies being applied to the modulator results in a better transmitted audio signal to noise ratio due to the highs being above the noise as much or more than the lows. So what is the original reason for Pre-emphasis? Bob Schmid covers that in the last section of this page.

Transmitters that employ a true FM modulator require a pre-emphasis circuit before the modulator fore the true FM modulator doesn't automatically pre emphasize the audio like a transmitter that uses a phase modulator. A separate circuit is not necessary for pre-emphasis in a transmitter that has a phase modulator because the phase modulator applies pre-emphasis to the transmitted audio as a function of the modulator itself.  In other words the phase modulator 'automatically' pre-emphasizes the applied audio.

The user's receiver de-emphasis circuitry takes the unnatural sounding pre-emphasized audio and turns it back into its original response. Pre-emphasized audio is however available directly from the audio demodulation or more commonly  the 'discriminator' circuitry.

In linking systems, many choose to eliminate the emphasis circuitry (on a whole) to allow better representation of retransmitted signals. Since the signal has already been pre-emphasized (by the user that is transmitting,) and since the receiver you are listening to takes care of the de-emphasis.... it doesn't need to be done over and over again.

Some loss of quality does exist in flat systems, but quality is better maintained by its use. A flat audio response system is one which has equal output deviation for the same input deviation, no matter what the applied audio frequency is.....within reason.

Reasonable audio frequency response would be from 50 cycles to about 3500 to 5000 cycles in a system not filtering the CTCSS tone. Audio response in a system that does filter the CTCSS would be around 250 to about 3500 to 5000 cycles. The upper cut off frequency would be determined mainly on acceptable use of available bandwidth....3500 on a 15 kHz 2 meter pair, or 5000 on a 25 kHz UHF pair.

Available Bandwidth:
Injecting discriminator audio back into an FM modulator without any limiting or low pass filtering is bad news, plain and simple.   On UHF, you may be able to get away with it without excessively bothering either of your adjacent channel neighbors, but on 2m, especially with 15 kHz spacing, you'd be asking for a lynching.

Without low pass filtering, all of the high frequency energy (hiss) that comes from the discriminator from a noisy user, if not low pass filtered, will deviate your transmitter in excess of 5 kHz, in addition to pushing the sidebands out further than they would be if the modulating frequencies were cut off at/about 3 kHz.   Do this to see what I am trying to get across.   Set your repeater up for 1:1 input to output ratio (like, put in a signal that is deviated 3 kHz by a 1 kHz tone, and set your Tx audio gain to get 3 kHz out of the transmitter). Now open your repeater receiver squelch wide open.   You should see your transmitter is now deviating somewhere around 8 or 9 kHz (presuming you have enough audio headroom through the controller).   Under this test condition, the combination of the excessive deviation and the lack of high frequency filtering will make your signal somewhere around 30 kHz wide instead of 16 kHz as it should be (very high modulation index.)   The only thing "limiting" the occupied bandwidth at that point is the dynamic range of the audio circuits in the controller & radio... and the natural high frequency rolloff of the discriminator's output noise.   Observe the occupied bandwidth on a spectrum analyzer to view this.

Obviously that's a worst case scenario, but the fact remains that you should have brick wall limiting at 5 kHz (a little lower at 15 kHz channel spacing), and possibly low pass filtering at 3 kHz (a little higher is OK on 25 kHz channels).

Summary:
A repeater can be built to utilize a flat audio response to maintain quality through the system. This is fairly easy in a system using a true FM modulator.   Usually some modifications to the controller are necessary to allow it, especially ones that have a speech synthesizer or a phone patch.   Systems using a phase modulator require de-emphasis before the modulator because of the unavoidable built-in pre-emphasis of the audio by this type of modulator.   For this reason... it is easier to utilize flat audio modifications and maintain quality audio in a system employing a true FM modulator in each transmitter.

Kevin Custer W3KKC, January 19 1999

The Reasoning behind Pre-emphasis in Narrow Band FM
By Bob Schmid WA9FBO of S-COM Industries

Yes, the FM broadcasting industry uses pre-emphasis and de-emphasis techniques to improve their signal-to-noise ratios. It's been correctly pointed out that audio frequencies below the breakpoint are transmitted flat, and audio frequencies above the breakpoint are transmitted pre-emphasized. (There have been other such "curves" used to tailor response, such as the RIAA curve in phonograph records, and the NAB curve in tape recording.)  But that isn't the original reason pre-emphasis and de-emphasis were used in narrowband radio. The early transmitters were PM (phase modulated), not FM, so they naturally had a 6 dB/octave pre-emphasis. PM became the standard modulation method. When FM transmitters came along, their audio had to be intentionally pre-emphasized to maintain compatibility with the PM transmitters already in service. In very early narrowband literature, you won't even find the terms "pre-emphasis" and "de-emphasis".  Engineers simply "rolled off" the audio in the receiver with a single pole filter because they had to defeat the PM modulator's characteristic "roll-up". The pre-emph and de-emph terms came from the broadcast people. (I wish the narrowband radio industry had better terms for these characteristics. Unlike the broadcasters with their middle-of-the-band breakpoint, in narrowband radio the breakpoints are outside the voice bandwidth.) So, de-emphasis has little to do with signal-to-noise radio and everything to do with making the response correct. If FM had always been used, there never would have been pre-emph or de-emph in narrowband radio.

We must recognize that early narrowband radio was intended for one transmitter, one receiver applications. This business of linking repeaters came much later. We pre-emph the audio to the FM transmitter to simulate PM, but must maintain a narrow bandwidth to be a good neighbor. So, we roll off the audio at, let's say, -3 dB at 3 kHz. If we hop through another similar system, the resulting audio is then down another 3 dB at 3 kHz, or a total of -6 dB at 3 kHz. This narrowing of the audio bandwidth is what everybody complains about in linked systems.

So, the popular answer is to eliminate de-emph and pre-emph in the repeater. Just feed the user's pre-emph'd audio from the repeater receiver's discriminator to the repeater transmitter after the pre-emph stage, thus bypassing the repeater's de-emph and pre-emph circuits, resulting in a "flat" repeater, right?. (Of course, you still have all those controller mods to make.) Everyone then assumes de-emph and pre-emph are evil!!! They must be, since the audio sounds better without them!

But from an engineering perspective, there is nothing inherently evil in pre-emph or de-emph. The transmitter still rolls off at 3 kHz. By feeding pre-emph'd audio to the transmitter, you are artificially increasing the amount of high-freq audio fed to it. You are "peaking" the transmitter so that it rolls up. You are effectively widening the bandwidth.

My response? First, let's at least admit that we are attempting to make narrowband systems into wideband systems. No bones about it.  You want a nice, high-fidelity linked system? Go after the transmitters. Replace their audio filters with high order, brick wall audio filters that allow wider bandwidth signals (at the expense of smaller guard bands between channels). Or use wide band links on higher bands. But if you want to continue defeating de-emph and pre-emph, at least admit that it's similar to putting an audio equalizer in line.

One last thing - - FM'ing crystals is really hard (they're nonlinear).   FM'ing a synthesized transmitter is easy.   So, if someone tells you he FM'ed his crystal controlled repeater transmitter with a few wiring changes and a capacitor, make him prove it.   What audio signal generator did he use to sweep the transmitter?   Was the audio level checked at each measuring frequency (the output level of some cheaper audio generators changes with the frequency)?   What receiver was used to produce the measured audio?   Is it's frequency response linear across the audio range?   Remember, the proof is in the pudding (lab grade test equipment).   If it's really an FM transmitter, the received audio will be at a constant amplitude regardless of frequency.   Anything else is modified PM.

Bob Schmid WA9FBO of S-COM Industries     January 19, 1999