Arithmetic, Geometric, and Harmonic Means in Signal Processing
In signal processing, different types of averages are used depending on what physical quantity is being averaged. The arithmetic mean, geometric mean, and harmonic mean are not interchangeable. Each one has a different meaning and is suitable for different signal-processing situations.
Arithmetic Mean
The arithmetic mean is the ordinary average
It is used when the quantities being averaged are additive.
Signal-processing applications
The arithmetic mean is commonly used for averaging time-domain samples, estimating DC components, reducing random noise, and calculating mean power.
For example, the DC value of a discrete signal can be estimated as
If the signal contains random noise with zero mean, averaging repeated measurements can reduce the noise component.
Another important example is mean-square power
The RMS value is then
So, the arithmetic mean is essential in RMS calculation, power estimation, ensemble averaging, Welch averaging, and noise reduction.
Typical examples
| Application | Meaning |
|---|
| DC estimation | Average signal level |
| RMS calculation | Average signal power |
| Ensemble averaging | Noise reduction |
| Welch PSD averaging | Stable spectral estimate |
| Moving average filter | Local smoothing |
In short, arithmetic mean is used when values contribute additively.
Geometric Mean
The geometric mean is defined as
or equivalently
It is used when the quantities are related by multiplication, ratios, or logarithmic scales.
Signal-processing applications
The geometric mean is useful when dealing with frequency ratios, logarithmic frequency axes, gains, spectral magnitudes, and octave-band analysis.
A representative example is the center frequency of a band. For a frequency band with lower frequency f1 and upper frequency f2 , the center frequency is usually defined by the geometric mean
This is because octave and fractional-octave bands are based on frequency ratios, not frequency differences.
For example, if a band extends from 707 Hz to 1414 Hz
The arithmetic midpoint would not correctly represent the center on a logarithmic frequency scale.
The geometric mean is also related to averaging on a dB scale. Since dB is logarithmic, averaging logarithmic quantities corresponds to a geometric mean in the linear domain.
Typical examples
| Application | Meaning |
|---|
| Octave-band center frequency | Center by frequency ratio |
| Log-frequency axis | Equal ratio spacing |
| Gain ratios | Multiplicative averaging |
| Spectral magnitude statistics | Log-domain averaging |
| Cepstral/log-spectrum analysis | Average in logarithmic domain |
In short, geometric mean is used when ratios or logarithmic spacing are important.
Harmonic Mean
The harmonic mean is defined as
It is used when the quantities being averaged appear in the denominator of a relationship, or when averaging rates, impedances, or reciprocal quantities.
Signal-processing applications
The harmonic mean is less common than the arithmetic mean, but it appears in important cases involving rates, bandwidth-related quantities, equivalent parameters, and reciprocal-domain averaging.
A useful interpretation is that the harmonic mean is suitable when the same “amount of work” or “amount of distance” is associated with different rates. In signal processing and acoustics, similar reciprocal relationships appear in impedance, compliance, conductance, and equivalent system parameters.
For example, in electrical and mechanical analog systems, parallel combinations often involve reciprocal addition. The equivalent resistance of parallel resistors is
For two equal-weighted values, the harmonic mean is closely related to this type of reciprocal averaging.
In spectral analysis, the harmonic mean can also appear when averaging quantities such as periods instead of frequencies. Since frequency is the reciprocal of period,
averaging periods and converting back to frequency is not the same as averaging frequencies directly. If the physically relevant quantity is period, the corresponding average frequency can involve a harmonic-type relationship.
For a ratio-type quantity,
the average rate should always be defined as
That is, the average rate is not necessarily the simple average of the individual rates. It is the ratio of the total numerator to the total denominator.
| Condition | Average rate |
|---|
| All ti are equal | Arithmetic mean |
| All xi are equal | Harmonic mean |
| Both xi and ti vary | Weighted average, ∑xi / ∑ti |
| Both xi and ti are equal | All rates are the same, so the average is identical |
Typical examples
| Application | Meaning |
|---|
| Averaging rates | Appropriate for reciprocal quantities |
| Period-frequency relationship | Frequency is reciprocal of period |
| Equivalent resistance/impedance concepts | Parallel reciprocal combination |
| Averaging time constants or rates | Depends on physical model |
| Effective parameter estimation | When reciprocal contribution dominates |
In short, harmonic mean is used when reciprocal quantities or rates are physically meaningful.
Fundamental Inequality
A (Arithmetic mean) ≥ G (Geometric mean) ≥ H (Harmonic mean)
Equality holds only when all values are identical, x1 = x2 = x3 = ... = xN
Comparison in Signal Processing
| Mean type | Mathematical form | Best suited for | Signal-processing example |
|---|
| Arithmetic mean | (∑ x i ) / N | Additive quantities | RMS power, DC value, ensemble averaging |
| Geometric mean | (∏ x i ) 1/N | Ratios and log scales | Octave-band center frequency, log-spectrum averaging |
| Harmonic mean | N / (∑ 1 / x i ) | Rates and reciprocals | Period-frequency relation, equivalent reciprocal parameters |
Practical Interpretation
The choice of mean depends on the physical meaning of the data.
If signal samples, powers, or independent contributions are being added, use the arithmetic mean.
If frequency ratios, gains, logarithmic spectra, or octave-band quantities are involved, use the geometric mean.
If rates, periods, reciprocal quantities, or equivalent parameters are involved, use the harmonic mean.
A compact way to remember this is
Arithmetic mean: additive quantities quantities, average in the linear domain
Geometric mean: multiplicative ratios, average in the logarithmic domain
Harmonic mean: reciprocal quantities, average in the reciprocal domain
Conclusion
Arithmetic, geometric, and harmonic means all have meaningful roles in signal processing, but they are used in different contexts. The arithmetic mean is used for additive quantities such as sample averages, power estimates, RMS values, and noise reduction. The geometric mean is used for ratios and logarithmic quantities such as octave-band center frequencies and log-spectrum analysis. The harmonic mean is used when reciprocal relationships are important, such as rates, periods, and equivalent reciprocal parameters.
Therefore, the correct mean is not chosen only by mathematics, but by the physical meaning of the signal-processing quantity being averaged.
Suggested Further Reading
You may also find these topics helpful:
Arithmetic, Geometric, and Harmonic Means in Signal Processing
In signal processing, different types of averages are used depending on what physical quantity is being averaged. The arithmetic mean, geometric mean, and harmonic mean are not interchangeable. Each one has a different meaning and is suitable for different signal-processing situations.
Arithmetic Mean
The arithmetic mean is the ordinary average
It is used when the quantities being averaged are additive.
Signal-processing applications
The arithmetic mean is commonly used for averaging time-domain samples, estimating DC components, reducing random noise, and calculating mean power.
For example, the DC value of a discrete signal can be estimated as
If the signal contains random noise with zero mean, averaging repeated measurements can reduce the noise component.
Another important example is mean-square power
The RMS value is then
So, the arithmetic mean is essential in RMS calculation, power estimation, ensemble averaging, Welch averaging, and noise reduction.
Typical examples
In short, arithmetic mean is used when values contribute additively.
Geometric Mean
The geometric mean is defined as
or equivalently
It is used when the quantities are related by multiplication, ratios, or logarithmic scales.
Signal-processing applications
The geometric mean is useful when dealing with frequency ratios, logarithmic frequency axes, gains, spectral magnitudes, and octave-band analysis.
A representative example is the center frequency of a band. For a frequency band with lower frequency f1 and upper frequency f2 , the center frequency is usually defined by the geometric mean
This is because octave and fractional-octave bands are based on frequency ratios, not frequency differences.
For example, if a band extends from 707 Hz to 1414 Hz
The arithmetic midpoint would not correctly represent the center on a logarithmic frequency scale.
The geometric mean is also related to averaging on a dB scale. Since dB is logarithmic, averaging logarithmic quantities corresponds to a geometric mean in the linear domain.
Typical examples
In short, geometric mean is used when ratios or logarithmic spacing are important.
Harmonic Mean
The harmonic mean is defined as
It is used when the quantities being averaged appear in the denominator of a relationship, or when averaging rates, impedances, or reciprocal quantities.
Signal-processing applications
The harmonic mean is less common than the arithmetic mean, but it appears in important cases involving rates, bandwidth-related quantities, equivalent parameters, and reciprocal-domain averaging.
A useful interpretation is that the harmonic mean is suitable when the same “amount of work” or “amount of distance” is associated with different rates. In signal processing and acoustics, similar reciprocal relationships appear in impedance, compliance, conductance, and equivalent system parameters.
For example, in electrical and mechanical analog systems, parallel combinations often involve reciprocal addition. The equivalent resistance of parallel resistors is
For two equal-weighted values, the harmonic mean is closely related to this type of reciprocal averaging.
In spectral analysis, the harmonic mean can also appear when averaging quantities such as periods instead of frequencies. Since frequency is the reciprocal of period,
averaging periods and converting back to frequency is not the same as averaging frequencies directly. If the physically relevant quantity is period, the corresponding average frequency can involve a harmonic-type relationship.
For a ratio-type quantity,
the average rate should always be defined as
That is, the average rate is not necessarily the simple average of the individual rates. It is the ratio of the total numerator to the total denominator.
Typical examples
In short, harmonic mean is used when reciprocal quantities or rates are physically meaningful.
Fundamental Inequality
A (Arithmetic mean) ≥ G (Geometric mean) ≥ H (Harmonic mean)
Equality holds only when all values are identical, x1 = x2 = x3 = ... = xN
Comparison in Signal Processing
Practical Interpretation
The choice of mean depends on the physical meaning of the data.
If signal samples, powers, or independent contributions are being added, use the arithmetic mean.
If frequency ratios, gains, logarithmic spectra, or octave-band quantities are involved, use the geometric mean.
If rates, periods, reciprocal quantities, or equivalent parameters are involved, use the harmonic mean.
A compact way to remember this is
Arithmetic mean: additive quantities quantities, average in the linear domain
Geometric mean: multiplicative ratios, average in the logarithmic domain
Harmonic mean: reciprocal quantities, average in the reciprocal domain
Conclusion
Arithmetic, geometric, and harmonic means all have meaningful roles in signal processing, but they are used in different contexts. The arithmetic mean is used for additive quantities such as sample averages, power estimates, RMS values, and noise reduction. The geometric mean is used for ratios and logarithmic quantities such as octave-band center frequencies and log-spectrum analysis. The harmonic mean is used when reciprocal relationships are important, such as rates, periods, and equivalent reciprocal parameters.
Therefore, the correct mean is not chosen only by mathematics, but by the physical meaning of the signal-processing quantity being averaged.
Suggested Further Reading
You may also find these topics helpful: