Dual-Polarized Rate-Splitting Multiple Access for mmWave Communications: A Practical Design and Prototype Implementation

• RSMA combines principles from SDMA and power-domain NOMA with rate splitting, and has been shown to outperform both in many scenarios while offering robustness to imperfect CSIT,,,,,.The introduction also notes that RSMA includes SDMA and power-domain NOMA as particular cases.Source: §I · p. 1
• NOMA exploits channel-strength diversity but scales less efficiently than SDMA as antenna numbers grow, and its reliance on SIC has posed practical deployment challenges,,.Source: §I · p. 1
• FR2 offers largely underused spectrum and larger bandwidths than sub-6 GHz, but propagation and penetration losses and increased hardware, channel-estimation, and synchronization challenges have slowed adoption.The upper-mid FR3 band is discussed as a trade-off between implementation complexity and spectrum resources.Source: §I · p. 1

A. RSMA Prototype Implementations

• Prior RSMA prototype implementations were based on software-defined radio.The supplied table caption identifies the table as a survey of RSMA prototype implementations.Source: A. RSMA Prototype Implementations · p. 1, Table I · p. 2

B. Motivation: mmWave and Dual-Polarized RSMA

• The transmitter uses channel-strength and polarization information rather than instantaneous phase CSIT, addressing a key channel-acquisition challenge identified for mmWave.Source: B. Motivation: mmWave and… · p. 2
• The design serves two users with a common stream and two private streams, assigning the common stream to one polarization and a superposition of private streams to the other.The messages for each user are split into common and private parts before the common messages are combined.Source: B. Motivation: mmWave and… · p. 2
• The practical design models finite input constellations, shaping transmitter superposition and constellation choices as well as receiver decoding strategies.The discussed strategies include SIC, JMLD, and treating discrete interference as noise.Source: B. Motivation: mmWave and… · p. 2

C. Contributions

• The contribution combines polarization diversity and RSMA for one transmitter and two users with dual-polarized antenna pairs, using discrete signaling and magnitude CSIT rather than instantaneous phase CSIT.The paper contrasts this practical model with prior generalized-signaling concepts.Source: C. Contributions · p. 2
• The paper compares SIC-plus-JMLD with JMLD-plus-TIN using mutual-information rate expressions and simulations across polarization, channel-strength, SNR, constellation, and power-allocation conditions.The reported comparison finds only a modest achievable-rate increase for SIC, at the cost of greater hardware complexity and receiver channel-estimation demands.Source: C. Contributions · p. 2

A. System Model

• The system model represents transmission and reception over orthogonal horizontal and vertical polarizations, with channel coefficients incorporating channel scaling and co- or cross-polarization effects.The formulation can also be extended to orthogonal circularly polarized antennas.Source: A. System Model · p. 3

B. Channel Model

• The channel model assumes block-static magnitudes and polarization coefficients but rapidly varying phase, so the transmitter knows channel magnitudes while receivers track phase for decoding.The mutual-information analysis uses an i.i.d. uniform phase assumption; the model assumes a dominant path and negligible multipath.Source: B. Channel Model · p. 3

A. Rate-Splitting Multiple Access, Transmitter

• Single-layer RSMA combines both users’ common parts into s0 and sends it on vertical polarization, while horizontally transmitting the superposition of private streams s1 and s2.The private-stream superposition forms an alphabet whose size is M1M2 when every resulting symbol is unique.Source: A. Rate-Splitting Multiple Access,… · p. 4

B. SIC Receiver

• SIC decodes the common stream on vertical polarization while treating private streams as noise, then subtracts its estimate before decoding on horizontal polarization.JMLD recovers each user's private-stream index from the estimated superposed symbol.Source: B. SIC Receiver · p. 5
• In practical SIC, the common stream is decoded and error-corrected over a block before private-stream decoding, reducing error propagation but increasing latency and hardware complexity.Soft decoding is also feasible.Source: B. SIC Receiver · p. 5
• For discrete signaling and finite block length, SIC flexibility is constrained by the need to reduce private-stream interference to the common stream; it also requires estimating the cross-polarization coefficient hiHV.SIC is optimal for Gaussian signaling and infinite block length when the common-stream achievable rate exceeds each private-stream rate before SIC.Source: B. SIC Receiver · p. 5

C. Non-SIC Receiver

• The non-SIC receiver applies TIN across polarizations and JMLD to private streams, requiring only hiVV and hiHH but leaving common-stream interference in private-stream decoding.The specific SIC and non-SIC decoders have the same asymptotic complexity, O(M0 + M1M2).Source: C. Non-SIC Receiver · p. 5

D. Achievable Rate and Symbol Error Rate

• Achievable rate is constellation-constrained capacity: the mutual information between channel input and output under discrete signaling, used to guide system design.The total user rate includes an allocation factor for the common stream, whose rate is limited by the lower common-stream rate across the two users.Source: D. Achievable Rate and Symbol Error Rate · p. 5
• SER evaluates the practical implementation, and because closed-form symbol error probability is challenging for arbitrary constellations, it is estimated through simulation or hardware measurements.The estimates use the decoders defined for the SIC and non-SIC receivers.Source: D. Achievable Rate and Symbol Error Rate · pp. 5–6
• The numerical comparison uses distinct rate notation for SIC and non-SIC receivers and is intended to assess the specific practical designs, not provide a general decoder comparison.The rate expressions are computed numerically in Appendix A.Source: D. Achievable Rate and Symbol Error Rate · pp. 5–6

IV. NUMERICAL RESULTS AND SYSTEM DESIGN

• Numerical evaluations of the rate formulas provide the design insights used to guide the practical implementation.The implementation is detailed in Section V.Source: §IV · p. 6
• The analysis focuses on power allocation, constellation selection, and receiver type.Source: §IV · p. 6
• The evaluations also consider imperfect CSIT robustness, alternative approaches, and scalability to more users.Source: §IV · p. 6

A. Numerical Results and System Design, Similar Channels

• At SXPD1 = SXPD2 = 20 dB, both receivers favor γ0 = 0.5 and γ1 = γ2 = 0.25, and achieve significant improvement over the Common-only strategy.The private-stream constellations have lower cardinality than the common stream and require less power for equal performance.Source: A. Numerical Results and System… · p. 6
• As SXPD falls to 10 dB and 3 dB, SIC favors more common-stream power; the Common-only strategy becomes comparable to SIC as polarization diversity becomes less exploitable.The non-SIC receiver degrades because, unlike perfect SIC, it retains common-stream interference during private-stream decoding.Source: A. Numerical Results and System… · p. 6
• With SXPD1 = 20 dB and SXPD2 = 10 dB, RSMA can disable user 2's private stream while retaining polarization-diversity benefit for user 1, and βi can adjust user-rate fairness.With appropriate power allocation, SIC and non-SIC again exhibit similar performance.Source: A. Numerical Results and System… · p. 6

B. Numerical Results and System Design, Different Channels

• With 10 dB channel-strength difference and poor SXPD, non-SIC with the common stream disabled achieves performance close to SIC by exploiting channel-strength diversity.The non-SIC setting uses 16-QAM private streams and γ2 = 0.95.Source: B. Numerical Results and System… · p. 6
• The achievable-rate plots compare SNR and SXPD for SIC and non-SIC receivers across power allocations, alongside a common-only strategy with private streams disabled.The common-only strategy uses 16-QAM in Fig. 1 and 256-QAM in Fig. 2.Source: Figure 1 · p. 7, Figure 2 · p. 7
• At SXPD1 = 20 dB and SXPD2 = 10 dB, setting γ2 = 0 disables user 2's private stream while user 1's private stream retains polarization diversity.The comparison uses a 16-QAM common stream and a 4-QAM private stream for user 1; SIC and non-SIC perform similarly with appropriate power allocation.Source: Figure 3 · p. 7

C. Receiver Selection

• Across the evaluated scenarios, the non-SIC receiver with the common-only strategy nearly matches SIC, except for Fig. 2b at moderate SXPD.The analysis assumes infinite blocklength and perfect SIC, while the practical implementation uses non-SIC; conclusions apply to the chosen design and two-user, single-antenna-pair scenario.Source: C. Receiver Selection · pp. 7–8

D. Magnitude and Polarization CSIT Considerations

• With fixed constellation and power allocation at 15 dB SNR, all tested configurations improve as polarization diversity increases, indicating limited impact from imperfect polarization CSIT.Precomputed look-up tables offer a trade-off between exploiting channel conditions, stability, and robustness, but do not guarantee optimality for specific conditions.Source: D. Magnitude and Polarization CSIT… · p. 8, Figure 5 · p. 8

E. Comparison with RSMA Full Polarization Diversity

• The proposed design achieves higher rates than transmitting replicas of all streams over both polarizations, even at moderate cross-polarization rejection levels.With one dual-polarized antenna pair, separate precoding for common and private streams is unavailable; replicas maximize inter-stream interference and require higher γ0 for reliable SIC.Source: E. Comparison with RSMA Full… · p. 9

F. Scalability to More Users

• When user count grows together with antenna count, efficient spatial multiplexing generally requires phase CSIT and phase coordination across transmit antennas.The paper points to prior study of increasing-antenna dual-polarized RSMA, including massive MIMO configurations.Source: F. Scalability to More Users · p. 9
• With a fixed pair of dual-polarized antennas, RSMA can serve users alongside orthogonal multiplexing or multiplex three users with disparate channel strengths using one common and an additional private stream.The three-user arrangement depends on efficient superposed-constellation design for the channel-strength disparity.Source: F. Scalability to More Users · p. 9
• Pairing users with similar polarization conditions achieves higher average rates than pairing users with different polarization conditions in the illustrated four-user scenario.The comparison uses identical channel strengths and adjusts constellation and power-allocation parameters for each pairing strategy.Source: Figure 7 · p. 9, F. Scalability to More Users · p. 10

A. General Considerations

• The prototype uses USRP X310 radios with external mixers and power amplifiers to operate at 25 GHz, with a 2.5 GHz intermediate frequency.The dual-linearly polarized horn antennas connect their two independent ports to separate USRP ports.Source: A. General Considerations · p. 10
• The 10 Msps sampling rate and 6.75 MHz channel bandwidth make the prototype a single-subchannel test rather than a total-throughput evaluation.The sampling rate is limited by over-the-air processing on laptop-connected USRPs; the paper uses spectral efficiency and SER as performance measures and leaves OFDM/SC-FDMA integration for future work.Source: A. General Considerations · p. 10

B. Transmitter

• The transmitter forms packets for the common and two private streams, maps bits to the selected constellations, and normalizes power allocation so γ0+γ1+γ2 = 1.The normalization keeps total transmitted power constant.Source: B. Transmitter · pp. 10–11, Figure 8 · p. 11

C. Receiver

• The receiver performs blind amplitude, timing, frequency, and phase synchronization before decoding streams and using packet CRC32 headers to estimate SER.A commercial implementation with periodic known sequences could reduce packet losses before synchronization convergence; those losses are excluded from SER measurements.Source: C. Receiver · p. 11

D. Final Deployment

• A single receiver emulates both users in turn, with fixed channel conditions during each SER acquisition making measurements expected to match simultaneous-receiver results.The receiver is reconfigured in software without changing the transmitter configuration.Source: D. Final Deployment · pp. 11–12, Figure 9 · p. 11

A. Polarization Measurements and General Considerations

• Anechoic-chamber measurements captured dual-polarized antenna components across azimuth and roll; at 25 GHz, measured χVV and χVH identify roll angles yielding SXPDV = 10 dB, enabling desired SXPD values for SER measurements.P1 and P2 receive approximately the same power level, while P3 is selected to demonstrate a 6 dB reduction in EVM and estimated SNRi relative to P1 or P2.Source: A. Polarization Measurements and… · p. 12

B. SER Measurements

• In Case (i), all three streams achieve similar SER performance versus SXPD when both users have similar channel conditions and SNR1 = SNR2 = 25 dB.The common stream uses 16-QAM, the private streams use 4-PAM with one rotated by 90 degrees, and γ0 = 0.5 with γ1 = γ2.Source: B. SER Measurements · p. 12
• In Case (ii), setting γ2 = 0 gives user 1 a private stream while the 16-QAM common stream with γ0 = 0.9 continues to serve both users.User 1's private stream uses 4-QAM with γ1 = 0.1, and both users have 25 dB SNR.Source: B. SER Measurements · p. 13
• Measurements and simulations agree well across the reported SER cases, with a slight deviation at low SXPD in Case (iii).The low-SXPD interference affects receiver digital processing blocks not modeled in simulations, which assume perfect channel estimation and ideal time and frequency synchronization.Source: B. SER Measurements · pp. 12–13, Figure 12 · p. 13
• In Case (iii), user 2 at 19 dB SNR can decode correctly by detecting the transmitted symbol's quadrant, illustrating exploitation of channel disparity through power allocation.User 1 is at 25 dB; all streams use 4-QAM, with γ0 = 0.3, γ2 = 0.6125, and γ1 = 0.0875.Source: B. SER Measurements · p. 13, Figure 13 · p. 13

VII. CONCLUSION

• The SIC/JMLD and non-SIC/JMLD-plus-TIN architectures show comparable performance when the transmitter knows the decoder architecture and adapts its transmission strategy.Source: §VII · p. 14
• SER measurements under different channel conditions and SXPD values agree well with software simulations, validating the prototype design and implementation.The prototype was tested at 25 GHz; the design may be adapted to the upper portion of the envisioned FR3 band, where the channel model remains valid.Source: §VII · p. 14

APPENDIX A MUTUAL INFORMATION ANALYSIS

• The rate analysis assumes infinite block length, equiprobable stream symbols, and fixed, known power-allocation values.For the private-stream superposition, it assumes no repeated symbols in the constellation.Source: Appendix A · p. 14

A. SIC with Perfect CSIR

• With perfect CSIR, the SIC analysis bounds common- and post-cancellation private-stream rates using mutual-information expressions conditioned on the channel.The private-stream calculation treats the other private stream as interference after the common stream is decoded and cancelled.Source: A. SIC with Perfect CSIR · p. 14

B. Non-SIC with Partial CSIR

• With partial CSIR, the non-SIC analysis models interference as AWGN to derive rate expressions for common and private streams.These expressions are conservative approximations rather than exact input-output mutual information, reflecting the decoder's interference model.Source: B. Non-SIC with Partial CSIR · p. 15