PaperlayerView exact revision on arXiv

Source-linked AI summary

Heterogeneous Semantic and Bit Communications: A Semi-NOMA Scheme

Xidong Mu, Yuanwei Liu, Li Guo, Naofal Al-Dhahir

arXiv:2205.02620v2cs.ITeess.SP

TL;DR

The paper addresses the lack of a tractable semantic-similarity expression for analyzing coexistence between semantic and bit communications. It fits semantic similarity with a generalized logistic function, proposes OMA, NOMA, and semi-NOMA for heterogeneous transmission, and reports semi-NOMA as the superior scheme because of its flexible transmission policy.

  • Problem

    Semantic-rate analysis is hindered by the lack of a closed-form expression for semantic similarity.

  • Method

    The paper uses generalized logistic regression and develops OMA, NOMA, and semi-NOMA frameworks with SvB-rate and power-region analysis.

  • Results

    Semi-NOMA is reported as superior to NOMA and OMA, with numerical examples validating the analysis.

  • Takeaways & Limitations

    Semi-NOMA offers a flexible transmission policy for heterogeneous semantic and bit communication.

  • Takeaways & Limitations

    The paper assumes each user can decode both semantic and bit streams.

Abstract

from arXiv · show

Multiple access (MA) design is investigated for facilitating the coexistence of the emerging semantic transmission and the conventional bit-based transmission in future networks. The semantic rate is considered for measuring the performance of the semantic transmission. However, a key challenge is that there is a lack of a closed-form expression for a key parameter, namely the semantic similarity, which characterizes the sentence similarity between an original sentence and the corresponding recovered sentence. To overcome this challenge, we propose a data regression method, where the semantic similarity is approximated by a generalized logistic function. Using the obtained tractable function, we propose a heterogeneous semantic and bit communication framework, where an access point simultaneously sends the semantic and bit streams to one semantics-interested user (S-user) and one bit-interested user (B-user). To realize this heterogeneous semantic and bit transmission in multi-user networks, three MA schemes are proposed, namely orthogonal multiple access (OMA), non-orthogonal multiple access (NOMA), and semi-NOMA. More specifically, the bit stream in semi-NOMA is split into two streams, one is transmitted with the semantic stream over the shared frequency sub-band and the other is transmitted over the separate orthogonal frequency sub-band. To study the fundamental performance limits of the three proposed MA schemes, the semantic-versus-bit (SvB) rate region and the power region are defined. An optimal resource allocation procedure is then derived for characterizing the boundary of the SvB rate region and the power region achieved by each MA scheme. The structures of the derived solutions demonstrate that semi-NOMA is superior to both NOMA and OMA given its highly flexible transmission policy. Our numerical results validate the analysis and show the superiority of semi-NOMA.

I. INTRODUCTION

The paper studies how to support semantic and conventional bit communications together under limited wireless resources, proposing semi-NOMA alongside OMA and NOMA. It characterizes their achievable trade-offs and finds semi-NOMA most flexible and effective.

  • The central design problem is enabling heterogeneous semantic and bit transmission for multi-user networks under limited radio resources.
  • The paper proposes an AP framework that simultaneously serves one semantics-interested user and one bit-interested user.
  • Three multiple-access schemes are considered: OMA, NOMA, and semi-NOMA, which splits the bit stream between shared and orthogonal frequency resources.
  • The study defines and characterizes the semantic-versus-bit rate region and power region for comparing the three schemes.
  • The NOMA SvB rate region generally does not contain the OMA region, while both are contained in the semi-NOMA SvB rate region.
  • Even with symmetric channel gains, OMA remains strictly suboptimal to semi-NOMA, and pairing a stronger-channel S-user with a weaker-channel B-user is preferred.

C. Organization and Notation

The paper introduces the semantic-rate setting, explains the DeepSC text-transmission model, and outlines the organization and notation used for the subsequent heterogeneous-access analysis.

  • II. SEMANTIC RATE AND APPROXIMATION: The semantic rate is introduced as a performance metric whose semantic-similarity component is approximated by a generalized logistic function.
  • A. Semantic Rate: DeepSC uses neural semantic and channel encoders and decoders to map text into semantic symbols and recover the original sentence.
  • A. Semantic Rate: The received DeepSC signal follows y = hx + n, after which channel and semantic decoders recover the sentence.
  • A. Semantic Rate: Semantic similarity compares the original and recovered sentences, depends on semantic symbols per word K and received SNR γ, and is written as ε(K, γ).
  • A. Semantic Rate: The semantic rate combines sentence semantic information, semantic-symbol allocation, and transmission bandwidth into an effective rate.

B. Semantic Rate Approximation

The paper approximates semantic similarity with generalized logistic regression to obtain a tractable semantic-rate model, then uses it to formulate heterogeneous MA schemes.

  • B. Semantic Rate Approximation: The semantic similarity lacks an explicit form, so the paper fits it against received SNR using data regression and a generalized logistic function.
  • B. Semantic Rate Approximation: For fixed K, semantic similarity is non-decreasing with γ and bounded between εmin and εmax.
  • B. Semantic Rate Approximation: The fitted logistic curve accurately approximates the measured semantic similarity and provides a tractable form for theoretical design analysis.
  • B. Semantic Rate Approximation: Increasing γ helps most in the low-SNR regime, while increasing K has a stronger effect when K is small.
  • III. SYSTEM MODEL: The proposed framework sends semantic and bit streams from one AP to an S-user and a B-user under limited bandwidth and power.
  • III. SYSTEM MODEL: OMA and NOMA use separate and fully shared frequency bands, respectively, while semi-NOMA splits the bit stream across both types of resources.

A. OMA

OMA transmits the semantic and bit streams over separate orthogonal frequency sub-bands, avoiding interference but limiting spectrum efficiency. Its SvB analysis allocates bandwidth and power between the streams while accounting for semantic similarity.

  • Transmission design: OMA sends the semantic and bit streams through two orthogonal frequency sub-bands, enabling interference-free transmission.The scheme allocates separate bandwidth and power resources to the two streams.
  • Advantages and limitation: OMA is easy to implement because the two streams are transmitted without interference, but its spectrum efficiency is limited.The interference-free design trades implementation simplicity for lower spectral efficiency.
  • Performance characterization: The OMA SvB rate region captures the semantic-bit trade-off produced by feasible bandwidth and power allocation under fixed radio resources.The region is defined over allocations satisfying the total bandwidth and transmit-power constraints.
  • Semantic metric: Semantic performance depends on both semantic rate and semantic similarity, so maximizing semantic rate alone may not guarantee effective transmission.The paper therefore incorporates the approximated semantic similarity metric into the semantic-rate analysis.
  • Power region: The OMA power region specifies the minimum transmit power required to achieve target semantic and bit rates under the available bandwidth.The analysis defines this region using feasible allocations and channel-dependent received SNRs.

C. Semi-NOMA

Semi-NOMA splits the bit stream across a shared semantic-bit sub-band and an orthogonal bit-only sub-band. This design combines NOMA-style sharing with interference-free bit transmission and includes OMA and NOMA as special cases.

  • Transmission design: Semi-NOMA splits the original bit stream into shared and orthogonal components, providing flexible heterogeneous transmission.One bit stream is paired with the semantic stream, while the other is sent separately to the B-user.
  • Transmission design: The shared sub-band uses NOMA-style superposition and SIC, whereas the orthogonal sub-band carries the other bit stream without interference.Bandwidths Wm and Wb satisfy Wm + Wb = W, with separate powers allocated to the semantic and two bit streams.
  • Rate characterization: The overall semi-NOMA bit rate combines the effective shared-sub-band rate with the bit rate from the orthogonal sub-band.The shared component accounts for decoding at both users, while the orthogonal component is transmitted separately.
  • Special cases: Semi-NOMA unifies NOMA and OMA as special cases through resource allocation: Wb = 0 yields NOMA, while pb,m = 0 yields OMA.Other transmission strategies can also be obtained through different bandwidth and power allocations.
  • Trade-off: Semi-NOMA offers greater transmission flexibility than OMA and NOMA but entails relatively high hardware complexity at both transmitter and receiver.The flexibility comes from jointly adjusting the two sub-band widths and the three transmit powers.

IV. SVB RATE REGION CHARACTERIZATION

The paper characterizes SvB-region boundaries by optimizing resource allocation for target semantic rates. The analysis identifies extreme operating points, feasibility conditions, and bandwidth-allocation behavior for the OMA-based characterization.

  • Boundary characterization: The SvB boundary is obtained by optimizing the achievable bit rate for each target semantic rate under feasible bandwidth and power allocations.The procedure characterizes all boundary points between the semantic- and bit-focused extremes.
  • Boundary characterization: The extreme boundary points allocate all radio resources to semantic transmission or all resources to bit transmission, producing (Smax, 0) and (0, Rmax).The bit-only extreme has zero semantic rate, while the semantic-only extreme achieves the maximum semantic rate.
  • Feasibility: Insufficient transmit power can shrink the achievable SvB region, so the characterization distinguishes power-sufficient and power-limited cases.The paper defines the corresponding objective value as zero when the optimization problem is infeasible.
  • Optimal allocation: The derived feasibility conditions bound the semantic bandwidth and simplify the optimization by making one constraint superfluous.The analysis uses upper and lower bandwidth bounds when solving the reduced problem.
  • Optimal allocation: When the required semantic rate is low, allocating less bandwidth to semantic transmission can maximize the bit rate.Further bandwidth reduction may require additional semantic power, so the objective is not generally monotonic in semantic bandwidth.

B. SvB Rate Region Characterization for NOMA

The NOMA SvB region is characterized by varying the power split while both streams share the full frequency band. The analysis compares NOMA with OMA and identifies NOMA’s spectrum-efficiency and SIC-complexity trade-off.

  • Boundary characterization: NOMA characterizes the SvB boundary by allocating semantic power and assigning the remaining transmit power to bit transmission over the shared band.The extreme semantic point uses all transmit power for semantic transmission, while intermediate points vary the semantic power.
  • Transmission design: NOMA uses the full frequency band for both streams, while SIC enables interference-free semantic decoding at the S-user.The S-user first decodes and removes the bit signal before decoding its semantic signal.
  • Comparison with OMA: NOMA’s SvB rate region necessarily contains the OMA SvB rate region under the paper’s considered conditions.This comparison addresses the relative performance of NOMA and OMA for heterogeneous semantic and bit transmission.

C. SvB Rate Region Characterization for semi-NOMA

Semi-NOMA characterizes the SvB rate-region boundary through bandwidth and power allocation, with OMA and NOMA included as special cases. Its flexible transmission policy makes its achievable region contain those of both alternatives.

  • The extreme points (Smax, 0) and (0, Rmax) are also achievable boundary points for semi-NOMA.The power-limited case requires a qualification for the semantic-only boundary point.
  • For a fixed bandwidth allocation and target semantic rate, the optimization enforces the semantic-rate constraint and total-power constraint while allocating power across shared and orthogonal sub-bands.The bit-rate objective is optimized using water-filling-based power allocation after accounting for the semantic-stream requirement.
  • All boundary points between (Smax, 0) and (0, Rmax) are characterized by optimizing semi-NOMA over the shared-bandwidth allocation Wm.The optimization evaluates each Wm in [0, W] and combines one-dimensional search with water-filling power allocation.
  • Semi-NOMA's SvB rate region always contains the regions achieved by OMA and NOMA because both are special cases of semi-NOMA.The result supports semi-NOMA's superiority through its more flexible transmission options.

V. POWER REGION CHARACTERIZATION

The power region is characterized by minimizing transmit power subject to semantic-rate and bit-rate requirements for OMA, NOMA, and semi-NOMA. Semi-NOMA uses bandwidth search and water-filling allocation to obtain its optimum.

  • The power-region characterization minimizes transmit power subject to the required semantic and bit rates for each proposed multiple-access scheme.The framework applies the optimization to OMA, NOMA, and semi-NOMA.
  • OMA and NOMA solutions are derived from their corresponding constrained optimization problems using one-dimensional bandwidth searches and the approximated semantic-similarity function.For NOMA, the semantic constraints are obtained by specializing the shared-bandwidth formulation.
  • Semi-NOMA's optimum is obtained by searching Wm over [0, W] and applying water-filling power allocation for the shared and orthogonal sub-bands.The allocation first determines the minimum semantic-stream power, then optimizes bit-stream power subject to non-negativity and the remaining power budget.
  • Since OMA and NOMA are special cases of semi-NOMA, the semi-NOMA minimum required power is no greater than that of either scheme.This structural inclusion follows directly from the feasible-policy relationship among the schemes.

A. SvB Rate Region Comparison

Numerical SvB-rate comparisons show that semi-NOMA consistently provides the broadest region across channel differences and encoding parameters. Its advantage comes from flexible resource sharing, while NOMA and OMA exhibit distinct regime-dependent limitations.

  • 1) Impact of Users’ Channels Differences: Semi-NOMA strictly contains the SvB rate regions of OMA and NOMA across all three channel-difference cases.The comparison covers S-user stronger, symmetric, and S-user weaker channel conditions.
  • 1) Impact of Users’ Channels Differences: NOMA's SvB region is restricted because its fully shared frequency band requires substantial semantic-stream power to satisfy the similarity constraint.Unlike OMA and semi-NOMA, NOMA cannot adjust the semantic-stream bandwidth independently.
  • 1) Impact of Users’ Channels Differences: In the high-semantic-rate regime, semi-NOMA and NOMA perform equally and outperform OMA, whereas OMA remains strictly suboptimal even for symmetric channels.This differs from conventional bit-only transmission, where OMA can match NOMA in symmetric channels.
  • 1) Impact of Users’ Channels Differences: With a fixed B-user channel, increasing the S-user channel gain enlarges the SvB region, with a more pronounced enlargement for semi-NOMA than OMA.The results suggest pairing a higher-gain S-user with a lower-gain B-user for higher performance.
  • 2) Impact of K: As K increases, the maximum semantic rate decreases, while semi-NOMA continues to contain the OMA and NOMA regions for each K.For semi-NOMA, the region at K = 10 is contained in that at K = 4, indicating limited benefit from further increasing K.

B. Power Region Comparison

Power-region experiments compare the schemes under target semantic and bit rates and varying encoding parameters. Semi-NOMA benefits from flexible resource allocation, while the relative power efficiency of OMA and NOMA depends on the operating regime.

  • S: NOMA's required power depends on the semantic-similarity constraint rather than the target semantic rate because its semantic rate is automatically achieved.This behavior differs from conventional bit-based communication, where NOMA is no worse than OMA.
  • S: OMA and NOMA are superior in low- and high-semantic-rate regimes, respectively, because orthogonal bandwidth helps OMA at low rates while spectrum sharing helps NOMA at high rates.The regime-specific advantage reflects the bandwidth and interference trade-offs of the two schemes.
  • Semi-NOMA is expected to achieve the best power performance because its flexible allocation combines shared and orthogonal frequency resources.The numerical discussion attributes its resource efficiency to partial spectrum sharing.
  • ε: As the semantic-similarity target ε increases, the required transmit power of all schemes increases.The supplied results also state that semi-NOMA is less sensitive to ε than the other schemes.
  • 3) Impact of K: The power-region trend can differ from the SvB-region trend because power characterization includes both power-sufficient and power-limited cases.The supplied footnote explicitly distinguishes this setting from the power-sufficient SvB comparison.

VII. CONCLUSIONS AND FUTURE WORK

The paper develops and evaluates heterogeneous semantic-and-bit transmission using OMA, NOMA, and semi-NOMA, finding semi-NOMA superior because of its flexible transmission policy. It also identifies user-pairing, semantic encoding/decoding, multimodal extensions, and multi-user coordination as important directions.

  • Conclusions: The study characterizes the semantic-versus-bit rate and power regions for OMA, NOMA, and semi-NOMA using the semantic rate as the semantic-transmission metric.The framework simultaneously serves one S-user and one B-user.
  • Conclusions: NOMA does not necessarily outperform OMA, while pairing a higher-gain S-user with a lower-gain B-user is preferable for semi-NOMA.This pairing is reported as guidance for maximizing semi-NOMA's performance gain.
  • Conclusions: Semi-NOMA outperforms OMA and NOMA in heterogeneous semantic-and-bit transmission because its flexible policy supports more transmission options.The scheme can encompass OMA and NOMA as special cases.
  • Conclusions: Optimizing the semantic encoding/decoding scheme is important for improving semantic transmission performance.The conclusion identifies semantic processing as an additional performance lever beyond multiple-access design.
  • Future work: The model is limited to one S-user and one B-user, leaving joint user pairing and resource allocation for multiple clusters as an open problem.A possible extension clusters one S-user with one B-user and applies OFDMA across clusters, but joint design requires further research.
  • Future work: Future work includes performance-limit characterization for multimodal semantic data and opportunistic policies that select semantic or bit transmission according to channel conditions.The paper also notes that each user is assumed to use only one transmission method.
Loading 2205.02620v2…