In traditional signal processing, we often treat signals as complex-valued or bipolar oscillations. However, a growing body of research across optical communications and compressed sensing reveals that when signals are physically constrained to be non-negative, the fundamental rules of information theory and recovery change. This "positivity constraint" is not merely a detail; it is a powerful mathematical prior that prevents destructive interference and alters the very capacity of communication channels.
The Physicality of Light: Why Positivity Matters
In optical wireless communication (OWC) and Visible Light Communication (VLC), information is transmitted via Intensity Modulation and Direct Detection (IM/DD). Because light intensity cannot be negative, the transmitted signal must be strictly real-valued and non-negative.
As Anas Chaaban and Zouheir Rezki demonstrate in their tutorial on IM/DD channels, this constraint invalidates the standard Shannon capacity formulas used in radio frequency (RF) systems. Applying traditional models leads to a significant overestimation of performance—often by as much as 3.6 dB—because they fail to account for the power efficiency lost to DC biases or signal clipping required to keep the waveform positive. To combat this, researchers like Jingjing Huang have developed specialized waveforms like X-OFDM (Two Types of Mixed Orthogonal Frequency Division Multiplexing), which cleverly manage sub-carriers to maintain non-negativity without the need for power-hungry DC offsets.
Positivity as a Recovery Superpower
In the realm of sparse recovery and compressed sensing, non-negativity acts as a safeguard against "signal cancellation." In general systems, a positive component and a negative component can sum to zero, making it impossible for a sensor to detect their presence. When signals are known to be non-negative, this ambiguity disappears.
- No Destructive Interference: Matthew Harper Langston and colleagues leverage this in A sparse multidimensional FFT for real positive vectors. By ensuring that aliased signals cannot cancel each other out, they achieve near-optimal complexity for the Sparse Fast Fourier Transform.
- Phase Transitions and Geometry: The mathematical "hardness" of recovering a signal is often determined by the geometry of the projected space. Jared Tanner showed in Counting the Faces of Randomly-Projected Hypercubes and Orthants that the "orthant" (the non-negative quadrant of space) has unique properties that dictate sharp phase transitions for signal recovery.
- Super-Resolution: In deconvolution, Quentin Denoyelle and Gabriel Peyré found that positive measures allow for "grid-free" recovery of spikes even when they are clustered far below the classical Rayleigh limit (Support Recovery for Sparse Deconvolution of Positive Measures).
The Cost of Complexity
While positivity simplifies some aspects of recovery, it complicates others. Many efficient "one-scan" algorithms and sketching methods—such as those discussed by Ping Li in Sparse Recovery with Very Sparse Compressed Counting—are currently restricted to non-negative signals. Extending these to mixed-sign signals often requires an order of magnitude more measurements or significantly more complex optimization frameworks, such as the $\ell_1$-BranchHull proposed by Alireza Aghasi (A convex program for bilinear inversion of sparse vectors).
Furthermore, in fields like seismic imaging, the requirement for non-negativity creates a bridge to Optimal Transport theory. Björn Engquist and Yunan Yang demonstrate that by treating signals as non-negative probability distributions, one can use the Wasserstein metric to bypass "cycle skipping" issues that plague traditional wave inversion (Optimal Transport Based Seismic Inversion).
Summary of Intuition
The research direction suggests that non-negativity is a fundamental physical and mathematical constraint that acts as a regularizer. In communication, it forces us to rethink capacity and waveform design to accommodate the physical reality of light. In sensing, it provides a "floor" that prevents signal cancellation, allowing for faster algorithms and higher resolution than would be possible with general signed data.