Polarization division multiplexing (PDM) offers a way to carry multiple optical data channels through a shared path by assigning each channel a different polarization angle. In a visible light communication (VLC) system, intensity carries the digital information while polarization distinguishes the parallel channels. A receiver then separates those components with polarization filters and converts the resulting light into electrical signals.
A CMOS-compatible receiver design evaluated through system-level simulation demonstrates reliable recovery for two- and three-channel configurations. The four-channel case exposes a central engineering trade-off: as more channels share the same optical signal, crosstalk can make some input bit combinations indistinguishable. For automation engineers and system designers assessing optical links, the results highlight both the integration potential and the importance of evaluating channel separation across the complete receiver.
How polarization division multiplexing separates channels
Conventional optical communications commonly distinguish channels by physical path or wavelength. PDM instead uses the polarization state of light. Each digital channel modulates an optical signal assigned a specific linear polarization angle, and the signals are combined into one incident beam. At the receiving end, a division-of-focal-plane (DoFP) polarimeter array uses filters oriented to the channel angles to separate the combined signal.
The design considered three channel arrangements:
- Two channels: polarization angles of 0° and 90°.
- Three channels: 0°, 60° and 120°.
- Four channels: 0°, 45°, 90° and 135°.
The receiver models each optical channel using a Stokes vector, a mathematical representation of light’s intensity and polarization properties. The combined input is represented as the sum of the active channels’ Stokes vectors. A Mueller matrix models how each polarization filter transforms that input, producing an output intensity for its associated detector.
This framework matters because real filters are not ideal. Their transmission characteristics, including imperfections, affect how much of each channel reaches each detector. The filter model was calibrated using measurements from fabricated devices, allowing the simulations to account for measured filter properties rather than assuming perfect separation.
From optical filter to digital output
Each filter output feeds an individual silicon photodiode and an electronic recovery channel. In the modeled signal path, the photodiode converts incident light into current. Bias circuitry and an amplifier condition the resulting electrical signal, and a comparator with hysteresis classifies it as a digital high or low.
This end-to-end view is important for system design. Optical separation alone does not determine whether a channel can be recovered: the filter output must also produce a usable signal for the photodiode and downstream circuit. The simulations therefore include both the optical filter behavior and the receiver electronics. The proposed receiver, including its optics, is described as compatible with standard CMOS fabrication processes.
Channel count is limited by crosstalk
Two and three channels
In the two-channel configuration, the orthogonal 0° and 90° channels produce filter outputs that can be distinguished using a fixed intensity threshold. The modeled receiver recovers both digital inputs. The three-channel arrangement also supports recovery with an appropriately selected threshold, but its noise margin is narrower because the channels interfere with one another at the filter outputs.
The three-channel system was also assessed under modeled filter-angle variation. The analysis considered a 5° fabrication error, including a configuration with filter angles of 5°, 55° and 125° instead of the nominal 0°, 60° and 120°. Although the filter output levels changed, the subsequent receiver signals were reported as almost identical in the simulations. Within this modeled case, the angle variation did not affect system performance.
Four channels
Adding a fourth channel, at 0°, 45°, 90° or 135°, makes the input combinations harder to distinguish. The modeled four-channel receiver cannot uniquely recover every possible four-bit pattern: only 10 of the 16 input combinations produce distinct output combinations. In particular, some different input patterns result in the same filter-output values, so a fixed threshold cannot reliably identify all channels’ bits.
This is a key design implication: increasing the number of polarization channels does not automatically increase the number of independently recoverable data streams. The paper identifies channel coding as a possible way to improve recovery in a four-channel system, but further work is needed to establish how coding could address the ambiguity. Larger channel counts are a research possibility, not a demonstrated outcome of the modeled system.
What the results mean for optical system design
For engineers evaluating compact optical links, the study illustrates a practical design approach: model the optical filters using measured characteristics, then evaluate their outputs together with the photodetectors and receiver circuitry. This helps expose limitations that may not be apparent when considering polarization filters in isolation.
The results also distinguish feasibility from scalability. Two- and three-channel PDM configurations showed successful recovery in simulation, and the three-channel design retained its reported behavior under the modeled filter-angle variation. The four-channel design, by contrast, demonstrated that crosstalk can prevent full data recovery even when each channel has a corresponding filter and receiver path.
Conclusion
Polarization division multiplexing can use polarization as an additional means of separating optical data channels over a common path. A CMOS-compatible DoFP receiver architecture provides a modeled route from polarization filtering through photodetection to digital recovery. The practical takeaway is to assess channel separability across the entire system: two and three channels were successfully modeled, while four channels encountered pattern ambiguity from crosstalk. Any move to higher channel counts will need to address that limitation, potentially through channel coding and further validation.