Complex Baseband Representation

Every modulated signal decomposes into two orthogonal components: the in-phase (I) component (real part) and the quadrature (Q) component (imaginary part). This decomposition is the physical basis for modern IQ modulators and demodulators.

• I-channel: Modulated by the cosine carrier, representing the real axis
• Q-channel: Modulated by the sine carrier, representing the imaginary axis
• Orthogonality: The 90° phase shift between carriers ensures the I and Q channels do not interfere
• Vector representation: The complex envelope s(t) = I(t) + jQ(t) fully describes the instantaneous amplitude and phase

Any real bandpass signal x(t) centered at carrier frequency f_c has a unique complex baseband equivalent x̃(t). The relationship is lossless and invertible:

x(t) = Re{ x̃(t) · e^(j2πf_ct) }

• No information loss: The complex baseband contains all modulation information
• Frequency shifting: Multiplication by e^(j2πf_ct) shifts the spectrum from baseband to passband
• Bandwidth preservation: The baseband signal bandwidth equals half the bandpass signal bandwidth
• Analytic signal connection: The complex envelope relates directly to the analytic signal via x_A(t) = x̃(t) · e^(j2πf_ct)

The complex baseband representation provides direct access to the instantaneous envelope and instantaneous phase of a modulated signal, which are critical features for modulation classification.

• Instantaneous amplitude: A(t) = |x̃(t)| = √(I²(t) + Q²(t)) — reveals amplitude modulation patterns
• Instantaneous phase: φ(t) = arg{x̃(t)} = atan2(Q(t), I(t)) — reveals phase modulation patterns
• Instantaneous frequency: f(t) = (1/2π) · dφ/dt — reveals frequency modulation patterns
• Feature extraction: Higher-order moments of A(t), φ(t), and f(t) serve as discriminative features for likelihood-based classifiers

The complex baseband maps directly to the constellation diagram, the geometric representation of all possible symbol states for a digital modulation scheme. Each symbol corresponds to a distinct complex value.

• QPSK: Four constellation points at {1+j, 1-j, -1+j, -1-j} normalized by 1/√2
• 16-QAM: Sixteen points arranged in a 4×4 grid with varying amplitudes
• Decision regions: Likelihood-based classifiers partition the complex plane into regions corresponding to each hypothesis
• Euclidean distance: The distance between received sample and constellation points drives the log-likelihood computation

The complex baseband representation enables single-sideband analysis, eliminating the redundant negative-frequency image present in real signals. This halves the required sampling rate and simplifies spectral analysis.

• Nyquist advantage: Complex sampling at rate B captures a bandwidth of B, while real sampling requires 2B
• No spectral redundancy: The complex baseband spectrum is not Hermitian symmetric
• Hilbert transform relationship: The Q component is the Hilbert transform of the I component for single-sideband signals
• Practical impact: IQ sampling at 100 MS/s captures 100 MHz of spectrum, critical for wideband modulation recognition

Under the Additive White Gaussian Noise (AWGN) assumption, the complex baseband representation yields elegant closed-form likelihood functions essential for ALRT, GLRT, and HLRT classifiers.

• Circularly symmetric noise: Complex AWGN has independent, equal-variance I and Q components
• PDF simplification: The complex Gaussian PDF is p(r|s) = (1/πσ²) · exp(-|r - s|²/σ²)
• Sufficient statistic: The matched filter output ∫ r(t)s*(t)dt is a sufficient statistic for detection
• Log-likelihood ratio: Reduces to a simple correlation operation: Re{∫ r(t)s*(t)dt} minus an energy bias term

The real and imaginary parts of the complex baseband signal, representing the cosine and sine carrier projections. Key properties:

• I-channel: Modulates the cosine carrier (0° phase)
• Q-channel: Modulates the sine carrier (90° phase)
• Together, they form a phasor with instantaneous amplitude and phase
• Any bandpass signal can be uniquely decomposed into I and Q components
• Example: QPSK encodes 2 bits per symbol by mapping to four distinct (I,Q) pairs: (1,1), (-1,1), (-1,-1), (1,-1)

The analytic signal is a complex-valued time-domain representation containing only positive frequencies, formed by adding the original signal to its Hilbert transform as the imaginary part. Critical aspects:

• Eliminates negative frequency components entirely
• The Hilbert transform applies a -90° phase shift to all frequency components
• Enables extraction of instantaneous amplitude, phase, and frequency
• Forms the mathematical bridge between real bandpass signals and complex baseband equivalents
• Essential for coherent demodulation in software-defined radios

A rotating vector in the complex plane that captures both magnitude and phase of a sinusoidal signal at a single frequency. In baseband analysis:

• The magnitude represents instantaneous signal envelope
• The angle represents instantaneous phase offset from the carrier
• Modulation schemes map directly to phasor trajectories:
  • QPSK: Four discrete phasor positions at 45°, 135°, 225°, 315°
  • 16-QAM: 16 phasor positions combining amplitude and phase shifts
• Constellation diagrams are static snapshots of phasor endpoints at symbol sampling instants

All bandpass filtering operations have an equivalent lowpass counterpart in the complex baseband domain, dramatically simplifying analysis:

• A bandpass filter centered at fc becomes a lowpass filter centered at 0 Hz
• The complex impulse response captures both magnitude shaping and phase distortion
• Enables efficient digital implementation using complex FIR/IIR structures
• Pulse shaping filters (raised cosine, root-raised cosine) are designed as lowpass prototypes
• Matched filtering for optimal detection is performed entirely in baseband

A mismatch between transmitter and receiver oscillators manifests as a rotating phasor in complex baseband, where the entire constellation spins at the offset frequency. Consequences:

• Static constellation points become circular trajectories
• Symbol decisions degrade rapidly without correction
• The rotation rate equals Δf = f_tx - f_rx
• Frequency offset estimation algorithms (e.g., M&M, Fitz) operate on the baseband phase trajectory
• Critical preprocessing step before any likelihood-based modulation classifier

In baseband, AWGN becomes a circularly symmetric complex Gaussian process where real and imaginary noise components are independent with equal variance. Properties:

• Noise power splits equally: σ²_I = σ²_Q = N₀/2
• The noise envelope follows a Rayleigh distribution
• The noise phase is uniformly distributed over [0, 2π)
• Likelihood functions for modulation classifiers are derived from this complex PDF
• The circular symmetry property simplifies CRLB and Fisher Information calculations