Signals: All-Pass and Minimum-Phase Systems

In signal processing, we often focus on how a system alters the magnitude of a signal's frequency components. However, the phase response is equally critical, determining the signal's shape in time. Understanding all-pass and minimum-phase systems provides the tools to characterize and manipulate this phase behavior, which is essential for applications from audio processing to communication channel equalization where timing and waveform integrity are paramount.

Magnitude, Phase, and the Transfer Function

A linear time-invariant (LTI) system is fundamentally characterized by its transfer function, $H(s)$ for continuous-time or $H(z)$ for discrete-time systems. This complex-valued function tells you how the system responds to each complex exponential input. You can always express it in polar form:

$$H(j\omega )=|H(j\omega )|e^{j\angle H(j\omega )}$$

Here, $|H(j\omega )|$ is the magnitude response and $\angle H(j\omega )$ is the phase response. While the magnitude response dictates how much each frequency component is amplified or attenuated, the phase response dictates how much each component is shifted in time. A distortion of the phase relationship between frequencies can drastically alter a signal's shape, even if the magnitudes are preserved. This leads us to two specialized classes of systems defined by their unique phase properties.

All-Pass Systems: The Phase Modifiers

An all-pass system is defined by a frequency response with unit magnitude at all frequencies: $|H_{ap}(j\omega )|=1$. It passes all frequencies with no amplification or attenuation, hence the name "all-pass." Its entire effect is on the phase (and consequently, the time delay) of the signal.

How is this possible? The structure of an all-pass transfer function provides the answer. For a continuous-time system, a first-order all-pass section has a transfer function of the form: $$H_{ap}(s)=\frac{s-a}{s+a},\text{where }Re(a)>0$$ Notice the pattern: a pole at $s=-a$ (in the left-half plane for stability) is paired with a zero at the mirror image location across the imaginary axis, at $s=+a$. This pole-zero mirroring across the $j\omega$-axis is the hallmark of an all-pass system. For every stable pole, there is a corresponding zero in the right-half plane. The magnitude cancels out to one for all $j\omega$, but the phase, given by $\angle H_{ap}(j\omega )=-2\arctan (\omega /a)$, is non-trivial and negative (lagging).

The primary utility of all-pass systems is phase equalization. If a desired system (like an audio filter) has a non-ideal phase response, you can cascade it with a specially designed all-pass filter. The all-pass filter's phase response is tailored to compensate for—or "equalize"—the unwanted phase distortion of the original system, making the overall phase response more linear without affecting the overall magnitude response.

Minimum-Phase Systems: The Efficient Responders

A stable, causal system is classified as minimum-phase if both its poles and zeros lie within the stable region (left-half of the s-plane for continuous-time, or inside the unit circle for discrete-time). This structural constraint has profound implications for its phase properties.

Among all causal and stable systems that have the identical magnitude response $|H(j\omega )|$, the minimum-phase system has two key properties:

1. It has the minimum possible phase lag (i.e., the least negative phase).
2. It has the minimum possible group delay. Group delay, defined as ${\tau }_g(\omega )=-\frac{d}{d\omega }\angle H(j\omega )$, measures the average time delay of a narrowband signal component at frequency $\omega$.

In essence, the minimum-phase version of a given magnitude response is the fastest-responding, least-delayed physical realization. Why does this matter? First, it is often the preferred design for filters where time-domain response speed is critical. Second, it forms a unique and reversible baseline: any system with a rational transfer function can be decomposed into a minimum-phase system followed by an all-pass system.

Decomposition into Minimum-Phase and All-Pass Components

This decomposition is a powerful analytical tool. Given any causal, stable transfer function $H(s)$, you can always factor it as: $$H(s)=H_{min}(s)\cdot H_{ap}(s)$$

Here’s the step-by-step process:

1. Identify all zeros of $H(s)$ that lie in the unstable region (right-half plane for continuous-time).
2. For each such non-minimum-phase zero at $s=z_i$ (with $Re(z_i)>0$), reflect it into the stable left-half plane at $s=-z_i^{\ast }$. The collection of these reflected zeros, combined with all original stable zeros and poles, defines the minimum-phase component $H_{min}(s)$. It has exactly the same magnitude response as the original system: $|H_{min}(j\omega )|=|H(j\omega )|$.
3. The all-pass component $H_{ap}(s)$ is constructed to account for the zero reflection. For each zero you reflected, $H_{ap}(s)$ includes a factor of the form $(s-z_i)/(s+z_i^{\ast })$. This all-pass section has the zeros in the unstable region and poles mirroring them in the stable region, giving it a unit magnitude.

This decomposition shows that any additional phase lag or group delay in a non-minimum-phase system, compared to its minimum-phase counterpart with the same gain, is attributable to the all-pass component tacked onto the end.

Application to Channel Equalization

The concepts of minimum-phase and all-pass systems converge in the critical task of channel equalization. Consider a communication channel (e.g., a telephone line or wireless link) with a transfer function $C(s)$ that distorts the transmitted signal.

The goal is to design a receiver filter, or equalizer $E(s)$, such that the overall response $C(s)E(s)$ is ideal. A common strategy is zero-forcing equalization, which aims for $C(s)E(s)=1$, implying $E(s)=1/C(s)$.

Here's where system type matters:

• If the channel $C(s)$ is minimum-phase, its inverse $1/C(s)$ is also stable and causal. You can directly design an equalizer that perfectly compensates for both the magnitude and phase distortion of the channel.
• If the channel $C(s)$ is non-minimum-phase (has zeros in the right-half plane), its direct inverse would be unstable (poles in the right-half plane) and unrealizable. The solution is to decompose $C(s)$ into $C_{min}(s)C_{ap}(s)$.

You then design the equalizer to invert only the minimum-phase part: $E(s)=1/C_{min}(s)$. This stable equalizer perfectly corrects the magnitude response and the minimum-phase portion of the phase distortion. The remaining all-pass factor $C_{ap}(s)$ introduces an uncompensated, but stable, phase shift. While perfect overall response isn't achievable, this method provides the best possible stable equalization of the magnitude, which is often the primary objective.

Common Pitfalls

1. Confusing "All-Pass" with "No Effect": A common error is thinking an all-pass filter doesn't change the signal because the magnitude spectrum is unchanged. This overlooks the fact that phase changes alter the time-domain waveform substantially. A simple all-pass filter can turn a symmetric pulse into a very asymmetric one, which matters in digital communications.
2. Equating Minimum-Phase with Linear Phase: A minimum-phase system has the least phase lag for its given magnitude, but that phase is not necessarily linear. A linear-phase system has constant group delay, which is a different, often desirable, property. A system can be minimum-phase, linear-phase, neither, or in rare special cases, both.
3. Misapplying the Decomposition: When decomposing a system, remember that the all-pass component must be stable and causal. This means its poles must be in the stable region. The decomposition rule reflects unstable zeros to stable locations; you never reflect poles.
4. Overlooking Stability in Equalization: The most critical pitfall in designing an inverse filter $1/C(s)$ is failing to check if $C(s)$ is minimum-phase. Attempting to directly invert a channel with a non-minimum-phase zero will result in an unstable, impractical equalizer. The decomposition method is the necessary workaround.

Summary

• All-pass systems have a frequency response with unit magnitude for all frequencies but a non-constant phase response. They are constructed via pole-zero mirroring and are used primarily for phase correction or shaping.
• Minimum-phase systems have all poles and zeros in the stable region of the s-plane. For a given magnitude response, they exhibit the minimum possible phase lag and group delay, representing the most efficient causal realization.
• Any stable, causal transfer function can be decomposed into the product of a minimum-phase system (containing the original magnitude response) and a stable all-pass system (containing the excess phase).
• In channel equalization, this decomposition is essential. Only the minimum-phase component of a channel's response can be stably and causally inverted. Equalizing this component corrects for magnitude distortion, while the remaining all-pass phase distortion must be tolerated or handled by other means.