Signals: Hilbert Transform and Analytic Signals

Understanding how to extract the underlying structure of a signal—its instantaneous amplitude and frequency—is crucial in fields like communications, radar, and vibration analysis. The Hilbert transform provides the mathematical toolset for this task, enabling the creation of a one-sided spectrum representation known as the analytic signal.

The Hilbert Transform: A 90-Degree Phase Shifter

At its heart, the Hilbert transform is a specific linear operation performed on a signal. It does not change the amplitude of frequency components; instead, it shifts the phase of every positive frequency component by -90° ($-\pi/2$ radians) and every negative frequency component by +90° ($+\pi/2$ radians). You can think of it as a phase shifter or a quadrature filter.

Mathematically, the Hilbert transform $\hat{x}(t)$ of a continuous-time real-valued signal $x(t)$ is defined by the convolution: $$\hat{x}(t) = \mathcal{H}\{x(t)\} = \frac{1}{\pi} \int_{-\infty}^{\infty} \frac{x(\tau)}{t - \tau} d\tau$$ This integral is a Cauchy principal value integral. In the frequency domain, the effect is elegantly simple. If $X(f)$ is the Fourier transform of $x(t)$, then the Fourier transform of its Hilbert transform, $\hat{X}(f)$, is given by: $$\hat{X}(f) = -j \cdot \text{sgn}(f) \cdot X(f)$$ where $\text{sgn}(f)$ is the sign function (1 for $f>0$, -1 for $f<0$, and 0 for $f=0$). The $-j$ term ($e^{-j\pi/2}$) implements the -90° shift for positive frequencies, while $+j$ handles the +90° shift for negative frequencies.

Constructing the Analytic Signal

The primary utility of the Hilbert transform is to generate the analytic signal $z(t)$. For a real signal $x(t)$, its analytic representation is a complex signal constructed as: $$z(t) = x(t) + j\hat{x}(t)$$ This formulation is profound. The real part is the original signal, and the imaginary part is its Hilbert transform, placing them in quadrature (90° out of phase). The most significant property of the analytic signal is its one-sided spectrum. In the frequency domain: $$Z(f) = \begin{cases} 2X(f), & \text{for } f > 0 \\ X(0), & \text{for } f = 0 \\ 0, & \text{for } f < 0 \end{cases}$$ All negative frequency components are suppressed. This compact representation is invaluable because it contains all the information of the original real signal but without the redundant conjugate-symmetric spectrum.

Computing Hilbert Transforms and Signal Envelopes

You often need to compute the Hilbert transform for standard signals. For a pure cosine, $\mathcal{H}\{\cos(2\pi f_0 t)\} = \sin(2\pi f_0 t)$. For a sine, $\mathcal{H}\{\sin(2\pi f_0 t)\} = -\cos(2\pi f_0 t)$. The Hilbert transform of a constant is zero. For a causal impulse response $h(t)$, a minimum-phase relationship exists: if $h(t)$ is causal (zero for $t<0$), its real and imaginary parts (its Fourier transform) form a Hilbert transform pair. This principle is central to the Kramers-Kronig relations in physics.

A key application is extracting the envelope $a(t)$ of a signal, which represents its instantaneous amplitude. For the analytic signal $z(t) = x(t) + j\hat{x}(t)$, the envelope is the magnitude: $$a(t) = |z(t)| = \sqrt{[x(t)]^2 + [\hat{x}(t)]^2}$$ For example, an amplitude-modulated signal $x(t) = m(t)\cos(2\pi f_c t)$ has an envelope approximately equal to $|m(t)|$ (for a slowly varying $m(t)$ compared to $f_c$). The instantaneous phase $\phi(t)$ is $\arg[z(t)] = \arctan(\hat{x}(t) / x(t))$.

Applications: AM Demodulation and Instantaneous Frequency

The analytic signal directly enables two powerful applications. First, in AM demodulation (envelope detection), you can recover a message signal $m(t)$ from a transmitted carrier without needing a synchronous local oscillator. By computing the analytic signal of the received AM wave $r(t) = [1 + m(t)]\cos(2\pi f_c t)$, its envelope $a(t) = |1 + m(t)|$ provides the demodulated output (followed by a DC block to remove the +1).

Second, the analytic signal allows for estimation of instantaneous frequency $f_i(t)$, which is the time derivative of the instantaneous phase (divided by $2\pi$): $$f_i(t) = \frac{1}{2\pi} \frac{d}{dt} \phi(t) = \frac{1}{2\pi} \frac{d}{dt} \arg[z(t)]$$ This is critical for analyzing frequency-modulated (FM) signals or non-stationary processes where frequency content changes over time, such as in chirp radar or biological signals. For a simple FM signal $x(t) = A\cos(2\pi f_c t + \beta\sin(2\pi f_m t))$, the instantaneous frequency calculated from its analytic signal will be $f_c + \beta f_m \cos(2\pi f_m t)$, correctly revealing the modulation.

Common Pitfalls

1. Treating it as a standard filter: The Hilbert transform's impulse response $1/(\pi t)$ is non-causal and infinite in length. In practice, you approximate it with a finite impulse response (FIR) filter, which introduces a delay and approximation error at band edges. It's not a simple bandpass filter.
2. Misapplying envelope detection: The envelope $a(t) = |z(t)|$ is only a meaningful representation of amplitude modulation if the carrier frequency is significantly higher than the bandwidth of the modulating message $m(t)$. Violating this narrowband assumption yields an envelope that doesn't correspond to $m(t)$.
3. Ignoring the DC component: The Hilbert transform of a signal with a DC component ($f=0$) is zero. The analytic signal's spectrum at $f=0$ remains $X(0)$, not $2X(0)$. Forgetting this can lead to errors in baseband signal processing.
4. Confusing phase unwrapping: Calculating instantaneous frequency requires differentiating the instantaneous phase $\phi(t)$. The $\arctan$ function yields a wrapped phase (usually between $-\pi$ and $\pi$). You must apply a phase unwrapping algorithm before differentiation to avoid jumps of $2\pi$ that create spikes in the estimated $f_i(t)$.

Summary

• The Hilbert transform is a -90° phase shifter for all frequency components of a signal, defined in the frequency domain by $\hat{X}(f) = -j \cdot \text{sgn}(f) \cdot X(f)$.
• Adding the original signal $x(t)$ and $j$ times its Hilbert transform creates the analytic signal $z(t)$, which possesses a one-sided spectrum containing no negative frequencies.
• The magnitude of the analytic signal provides the envelope $a(t)$, enabling simple AM demodulation, while its phase argument allows calculation of instantaneous frequency $f_i(t)$.
• Successful application requires adherence to the narrowband assumption for envelope detection and careful phase unwrapping for instantaneous frequency estimation.
• The Hilbert transform establishes a fundamental minimum-phase relationship between the real and imaginary parts of the Fourier transform for causal systems.