The Effects of AM and FM Modulation

The Effects of AM and FM Modulation

The ultimate goal of professional audio recorder is to capture and recreate audio waveforms without introducing any changes to the original waveforms.  Unfortunately, the analog magnetic recording process has limitations that do create changes in the reproduced waveform.  Some of these changes, such as frequency response anomalies, are of a linear nature that can be easily removed with corrective linear devices such as equalizers.  Other phenomena, such as harmonic and intermodulation distortions due to device nonlinearities, add new frequency components that are created strictly by the waveform passing through the system.  A third degradation is additions to the signal which are not at all related to the signal such as the random noise of the amplifiers and other components in the signal path.  All of these mechanisms are at work within the electronic circuits that amplify and process audio signals.

Another serious degradation known as modulation products is present in analog recording devices.  Unlike the previously mentioned types, modulation involves not only the desired signal, but also an external disturbance that is combined with the signal through a non-linear multiplying process.  In an analog magnetic recorder these modulation products are produced by tape speed variations known as wow and flutter, and irregularities in tape sensitivity known as tape dropouts and modulation noise.

Both of these modulation mechanisms create spurious frequency components that depend upon not just the audio signal, but also the magnitude and frequency of the external phenomenon creating the modulation.  For example, an eccentric capstan rotating at 10 revolutions per second in a tape recorder may produce speed variations at a rate of 10 Hz.  The amplitude of the variation is a function of the amount of eccentricity.  This speed variation will create new "ghost spectrum" components in the recorder output which are similar to the spectrum of the incoming audio signal, but shifted 10 Hz up and 10 Hz down on the frequency scale.  If a 1 kHz signal is entering the recorder, the output will have not only 1 kHz, but also .990 kHz and 1.01 kHz at reduced levels.  In fact, due to the complicated nature of the frequency modulation produced by wow and flutter, sideband frequency components will also be present at diminishing amplitudes at +20 Hz, + 30 Hz and so on from the original signal for every frequency component in the incoming signal.

                        AM and FM spectrum samples

Unlike the harmonic frequencies produced by second or third harmonic distortion, these new frequency components are not at all musically related to the source.  The speed of the capstan rotation has no relation to Bach or rock and roll.  And since the sidebands straddle every frequency component in the original signal, a simple filter cannot be used to undo the damage.

Flutter and tape dropouts are two different types of modulation.  Flutter is frequency modulation (FM) which results from changes in the speed of the tape.  Dropouts produce amplitude modulations (AM) similar to rapid gain changes.  The mathematics of frequency and amplitude modulation are very different, but both produce sideband pairs. [1]   The frequency span of both modulations is quite large, but can be broken down to characteristic groups.

Flutter is subdivided into three types - wow, flutter and scrape flutter.  Both wow and flutter are created by speed or tension variations created by mechanical imperfections in the tape transport.  Typical sources include eccentricities of rotating components, bad bearings, torque or speed variations in motors and variations in friction or drag. 

Speed variations at slow rates are perceived as pitch changes in the source material.  For a pure tone the effect is similar to pronouncing "wowowowow," hence the term wow.  Listeners are able to easily identify wow because natural phenomena such as doppler shifts create similar frequency shifts at low rates

Above approximately 5 Hz the listener no longer hears the basic tone shifting in pitch.  The perception changes to a broadening of the tone caused by the FM sidebands.  Although the flutter components are readily apparent, untrained listeners have difficulty describing the effect.

As the flutter rate increases further, the sidebands move so far away from the musical tones that the relationship between the tones and the sidebands is not perceived.  The sidebands create a veil around all of the music that degrades the crispness and transparency of the music.  Mechanically induced flutter in reel-to-reel audio recorders usually does not exceed a few hundred Hertz because the masses of the mechanical components serve as lowpass filters for mechanical vibration.  (The rotary heads of video and some digital machines can cause flutter components up to 1 kHz.) 

Unfortunately, scrape flutter does not benefit from the lowpass filtering effects of rotating components.  Scrape flutter is caused by a "stick/slip" mechanism that is more complicated in nature than the flutter generated by rotating components.  As the tape slides over the stationary surfaces of the guides and heads, vibrations are excited within the tape.  Just as in the bowing of a violin string, the vibration is started by the sliding of the tape over the stationary surfaces.  The vibration frequency is determined by the "tuning" of the tape strand.  Since the tape is a tensioned elastic member - stretched plastic - which has a density, length, and stiffness or modulus of elasticity, the tape will have a resonant vibration frequency.  For the tape paths of conventional recorders the resonance will range from as low as 3 kHz to as high as 10 kHz.  The peak is typically quite broad due to the somewhat random nature of the stick/slip phenomenon, just as a new violin student produces less than a pure tone at his first lesson.

The perception of scrape flutter is once again a seemingly random "modulation noise" which is behind the signal.  Modulation noises in general are characterized by a complex relationship between the desired signal and the unwanted modulating phenomenon.  First, when there is no signal, there is no noise.  The modulation noise will cease to exist during quiet passages, making measurement difficult.  Second, the nature of the noise depends upon the modulation phenomenon.  The noise sidebands depend upon the type of modulation such as amplitude, frequency, or phase modulation and the rate and amplitude of the modulation.  Typical modulations and their causes are:

Flutter                                                  Frequency modulation due to tape velocity variations

Tape dropouts and granularity noise      Amplitude modulation due to magnetic particle                                                                          inhomogenieties

Clock jitter in digital systems                 Phase modulation due to timing variations (which are very                                                                      similar to flutter, but primarily caused by the nature of the                                                                  sequence of ones and zeros in the data stream)

A few "rules of thumb" can be applied to scrape flutter:

1.  The length of the vibrating tape is the distance between massive rolling surfaces that are tightly coupled to the tape.  The rollers act as vibration energy absorbers or attenuators.  Do not expect light rollers made of plastic or aluminum to effectively damp scrape flutter.  If the tape tension or the wrap angle around the roller is inadequate to tightly couple the tape to the roller, the energy of the tape will not be absorbed into the roller.

2.  The longer the span, the lower the scrape flutter frequency and the higher the Q of the resonance.  Short unsupported spans raise the resonant frequency to allow the vibrations to be attenuated by mechanical losses within the tape.

3.  All mechanical elements such as guides and heads within the tape path must be rigid so that they do not vibrate at their own resonant frequencies.  Heads mounted on springs can be troublesome.

4.  Reduce tape tension and surface roughness to reduce the amplitude of the vibrations.  Smoothness and lubrication of the tape surface can vary drastically between tape types and manufacturers.

5.  If anything in the tape path is squealing, the resulting scrape flutter will be extremely high.  Tape that squeals as it passes through the capstan and pinch roller is a prime example.

6.  Adding any sliding surface to the tape path with increase scrape flutter.  Timecode heads or extra heads for multiple formats will degrade the scrape flutter performance.

7.  Avoid any scraping contact with the edge of the tape.  The slit edge is very rough and gets rougher with use.

8.  Changing the tape type or thickness will change the resonance of the tape.

9.  Scrape flutter is just as big a problem on 2" tape decks as it is on narrow tape formats.

                        Sample layout and calculation

The Perception of Flutter

Weighted flutter

Scrape flutter   

The ear/brain's ability to perceive sidebands

                        Low frequency Vs high frequency

                        Cyclic Vs random

                        Chart of ranges and perceptions

Flutter Measurement

In theory, flutter measurement is quite simple.  Simply reproduce a tape that has the appropriate test frequency on the deck to be tested and measure the amount of frequency modulation produced by speed and tension variations.  The first problem is picking the frequency of the test tone.  Just as with the sampling theorem of digital audio, the carrier frequency in a modulation system must be at least twice the frequency of the maximum information desired from the system.  In simple terms, the lower sideband, which is at the difference between the carrier and the maximum information frequency, must be a positive frequency.  The system must also pass the upper sideband, which is the sum of the two frequencies.

                        Block diagram of flutter measurement

For mechanical flutter components, the upper flutter limit is typically 1 kHz.  As mentioned earlier, the mass limitations of rotating elements, even on rotary head transports, generally keep vibrations below this limit.  Test frequencies of 3000 Hz and 3150 Hz can provide adequate bandwidth for these mechanical flutter components.  Various international standards bodies, including the NAB, IEC and JIS, have set standards for test instruments for mechanical flutter evaluation.  Standard measurement bandwidths are usually limited to 200 Hz, but a few test instruments have extended this bandwidth to 1200 Hz.

The lower limit is typically .5 Hz.  Below this frequency, speed variations are considered to be "speed drift" which can be measured with a frequency counter or a DC-coupled FM demodulator.  Most newer flutter measurement instruments do not have a drift function.

Unfortunately, as mentioned previously, the range of scrape flutter does not begin until approximately 3 kHz.  These frequencies are well outside the measurement capability of an instrument which has a test tone or carrier at this frequency!  A much higher test frequency is required.  None of the international standards organizations has yet addressed the topic of scrape flutter measurement.

Since most audio recorders are expected to provide reasonable record/reproduce response up to 18 kHz [2] , the upper limit of carrier plus measurement bandwidth can be chosen as 18 kHz.  Picking a 12.5 kHz carrier and a 5 kHz bandwidth provides a reasonable compromise requiring 17.5 kHz as the maximum frequency.  The lowest sideband of interest is 7.5 kHz, yielding a rolloff band from 5 kHz to 7.5 kHz to avoid severe aliasing. [3]

Audio Precision and Altair Electronics offer flutter measurement equipment that measure "High Band" flutter using a 12.5 kHz test tone.  The High Band measurement bandwidth can be selected either to extend to 5 kHz for scrape flutter measurements or to be limited to a maximum of 200 Hz for conventional rotational flutter measurements consistent with the published standards.  Moving the test tone up from 3 kHz to 12.5 kHz does not affect the low frequency flutter readings in any way.  The only result is the ability to extend the measurement bandwidth to include the scrape flutter components.  (The recorder must have adequate frequency response to meet the requirements of the previous paragraph.)

Now that a frequency for the test tone has been selected, the next task is to record a test tape that has no flutter from the record process.  Since any recorder, including the test tape recorder, will have flutter, a flutter-free tape is impossible.  In many cases, however, a purchased "flutter test tape" can have insignificant rotational flutter compared to the machine being tested.  (Beware that scrape flutter can be very high on these flutter test tapes.)

For high quality studio recorders with very low flutter, no suitable test tape is commercially available.  In this case the recorder must be used to both record and reproduce the test tape.  But do not fall into a common trap!  The simultaneous record/play mode can give false readings.   If the transit time from the record head to the play head is a multiple of the period of any flutter component, that flutter component will be canceled out. 

For example, consider a recorder with a capstan that has a circumference equal to the distance between the record and reproduce head gaps.  If the capstan is slightly eccentric, the speed of the tape will vary cyclically with each revolution of the capstan.  However, because the tape repeats the speed cycle as a recorded section reaches the reproduce head, the reproduced signal will have no apparent speed error.  If the tape is rewound and replayed, the eccentricity of the capstan may come up in any arbitrary relationship to the recorded error.  For some cases the errors will cancel as during recording.  In other cases, however, the errors will add, giving twice the flutter caused during recording.

The moral to this story is to rewind and replay the tape several times.  Even take the time to manually reposition the rotating components by hand to assure random orientation.  The reported value for the transport is the average for the series of tests.

(Always test the oscillator for low FM content before recording a test tape.  The "Input Monitor" mode of the recorder will feed the oscillator signal directly to the flutter meter.  Function generators are quite notorious for having significant amounts of flutter caused by AC mains ripple on the power rails.  Most general-purpose applications for function generators do not require the .01% or less FM noise floor that is necessary for testing high quality studio machines.)

The Measurement Device

Typical flutter specifications are quoted as one or two numbers that are supposed to tell everything important about the performance of the recorder.  Unfortunately, this is a gross simplification of reality.  Different parts of the world utilize different standards for the measurement bandwidth and the dynamics of the metering device. 

The U.S. standard arose from the National Association of Broadcasters (NAB) specification for flutter measurement.  The flutter is to be measured from .5 to 200 Hz using an average responding, RMS calibrated meter with meter dynamics as stated in the specification.

                        Filter curves

                        Table of comparison

The European standard from CCIR uses the same measurement bandwidth, but employs a quasi-peak meter that captures and stretches peak readings rather than average readings.  Since the meter is calibrated in peak values, the same machine will have higher CCIR values than NAB values.

The Japanese standard is based upon a true RMS responding meter.

All three standards include the weighting curve that focuses on components in the 3 Hz region.

Spectral Analysis

Spectral analysis for preventive maintenance and troubleshooting

            Obtaining stable representative spectra


                        Average or peak

Recommendations for standard methods of reporting flutter

            Sample reports

The sources of mechanical flutter

            Rotational speed variations

                        Motor speed and torque variations


                        Ball bearing calculations

            Calculating rates

                        Frequency = Distance/Speed

                        Example calculations

            Other sources


                        Ball bearings



                        Variable drag

            Typical scrape flutter spectra

                        Example spectra

Measurement techniques

            Levels, tape types, and bandwidth-related errors

            Interpreting demodulated products on the 'scope

            Finding the source of flutter

[1] For the math purist, the lower sideband of narrow band FM is reversed in polarity from the AM lower sideband.  This relationship can be used to construct a demodulator which can separate the FM and AM components.  In the general case, the magnitude of the FM sidebands decay with Bessel function coefficients.

[2] Slow speed decks such as minicassette and home video formats such as VHS require lower test frequencies.

[3] In practical designs the baseband lowpass filter is not as steep as antialias filters used in digital systems.  Some aliasing above 6.25 kHz may result, but the total error is small.

Back to top