Hilbert Transform

The defining property of the Hilbert transform is its ability to shift the phase of every positive frequency component by -π/2 radians (-90°) and every negative frequency component by +π/2 radians (+90°). The amplitude spectrum remains completely unchanged. This phase shift is what enables the separation of a signal into its in-phase and quadrature components, forming the foundation of the analytic signal representation.

In the frequency domain, the Hilbert transform acts as an ideal all-pass filter with a piecewise constant phase response. Its transfer function H(f) is defined as:

• H(f) = -j for f > 0 (positive frequencies)
• H(f) = 0 at f = 0 (DC component removed)
• H(f) = +j for f < 0 (negative frequencies) This means the magnitude response is unity across all frequencies, while the phase jumps by π radians at the origin.

The impulse response of the ideal Hilbert transformer is h(t) = 1/(πt), a two-sided, infinite-duration function. This makes the ideal Hilbert transform non-causal—the output at any time t depends on both past and future input values. In practice, this necessitates the use of finite impulse response (FIR) approximations with a processing delay, often designed using the Parks-McClellan algorithm or windowing methods.

The Hilbert transform is mathematically defined as the convolution of a real signal x(t) with the kernel 1/(πt). This is expressed as a Cauchy principal value integral:

codeCopy
H[x(t)] = (1/π) * PV ∫ [x(τ)/(t-τ)] dτ

The principal value is required because the integrand has a non-integrable singularity at τ = t. This convolution operation is what produces the 90-degree phase shift in the frequency domain.

The Hilbert transform's most critical application is constructing the analytic signal z(t) = x(t) + j·H[x(t)], where x(t) is the original real signal and H[x(t)] is its Hilbert transform. This complex-valued representation:

• Suppresses negative frequency components entirely
• Enables extraction of instantaneous amplitude (envelope) as |z(t)|
• Enables extraction of instantaneous phase as arg(z(t))
• Enables computation of instantaneous frequency as the derivative of the unwrapped phase

A key property for signal processing is Bedrosian's theorem, which states that if a(t) is a low-frequency signal and c(t) is a high-frequency carrier with non-overlapping spectra, then:

codeCopy
H[a(t) · c(t)] = a(t) · H[c(t)]

This allows the Hilbert transform to be applied directly to modulated signals, treating the slowly-varying envelope as a constant during the phase-shift operation. This is essential for envelope detection in communication systems.