OPV2V-R: Iterative Receiver for V2V Channels

• The paper introduces OPV2V-R as an iterative receiver that combines orthogonal precoding, PIC, and symbol-wise ML detection in OFDM-based V2V links to fully exploit time-frequency diversity.
• It demonstrates that constant-modulus orthonormal bases like DSFT and WHT yield identical per-symbol performance in doubly selective channels, simplifying basis selection.
• The architecture integrates iterative channel estimation and Wiener filtering, achieving up to 4.8 dB SNR gain at high mobility while balancing complexity and reliability.

OPV2V-R denotes an iterative receiver architecture for orthogonal precoding in doubly selective vehicle-to-vehicle channels, developed for OFDM links under IEEE 802.11p-like parameters and documented in “Iterative Detection for Orthogonal Precoding in Doubly Selective Channels” (Zemen et al., 2017). The design combines orthogonal precoding (OP), iterative parallel interference cancellation (PIC), symbol-wise maximum-likelihood detection, and iterative channel estimation in order to exploit full time–frequency diversity. In the formulation considered, OP spreads each data symbol across the entire time–frequency grid of an OFDM frame by means of a complete set of two-dimensional orthonormal sequences, while the receiver uses iterative processing to recover a scalar per-symbol model even though the physical channel is simultaneously delay- and Doppler-selective.

1. Definition and signal model

OPV2V-R is built on an OFDM system with a cyclic prefix long enough to avoid inter-symbol interference and with inter-carrier interference small enough to be neglected for IEEE 802.11p parameters (Zemen et al., 2017). The transmitter is assumed to have no channel-state information, the transmit window is rectangular, and the basis sequences are orthonormal over the $MN$-point time–frequency grid.

Given QAM data symbols $b_{p,n}$ on a transform-domain grid, orthogonal precoding maps them to the time–frequency grid as

$$d_{q,m}=\sum_{p=0}^{N-1}\sum_{n=0}^{M-1} b_{p,n}\, s^{p,n}_{q,m}.$$

With vectorization,

$\boldsymbol{d}=\boldsymbol{S}\,\boldsymbol{b},$

where $\boldsymbol{S}$ is the $MN\times MN$ matrix whose columns are the vectorized basis functions (Zemen et al., 2017).

After OFDM demodulation, the time–frequency-domain input–output relation is

$$\boldsymbol{y}=\operatorname{diag}(\boldsymbol{g})\,\boldsymbol{S}\,\boldsymbol{b}+\boldsymbol{n},$$

with $\boldsymbol{g}$ the sampled time-varying frequency response and $\boldsymbol{n}$ AWGN with entries $n_{q,m}\sim\mathcal{CN}(0,\sigma_n^2)$ (Zemen et al., 2017). Defining

$b_{p,n}$0

the receiver observes a channel-distorted precoding matrix. Because the channel multiplies $b_{p,n}$1, the transmit-side orthonormality of the sequences is generally lost at the receiver.

The underlying doubly selective baseband channel is represented by

$b_{p,n}$2

with $b_{p,n}$3 so that OFDM demodulation yields per-symbol, per-subcarrier samples $b_{p,n}$4 with negligible ICI (Zemen et al., 2017). A geometry-based simulation model for the sampled time–frequency response is

$b_{p,n}$5

where $b_{p,n}$6 is normalized Doppler, $b_{p,n}$7 normalized delay, and $b_{p,n}$8 path weights.

2. Orthogonal precoding bases and constant-modulus equivalence

A central component of OPV2V-R is the use of a complete orthonormal basis over the two-dimensional time–frequency grid. The discrete symplectic Fourier transform (DSFT) basis is a constant-modulus orthonormal basis with elements

$b_{p,n}$9

Its entries satisfy

$$d_{q,m}=\sum_{p=0}^{N-1}\sum_{n=0}^{M-1} b_{p,n}\, s^{p,n}_{q,m}.$$0

so each data symbol is uniformly spread over all $$d_{q,m}=\sum_{p=0}^{N-1}\sum_{n=0}^{M-1} b_{p,n}\, s^{p,n}_{q,m}.$$1 grid points (Zemen et al., 2017).

Walsh–Hadamard transform (WHT) sequences constitute another constant-modulus orthonormal basis. Their columns also satisfy $$d_{q,m}=\sum_{p=0}^{N-1}\sum_{n=0}^{M-1} b_{p,n}\, s^{p,n}_{q,m}.$$2 together with orthonormality (Zemen et al., 2017). The common orthogonality condition is

$$d_{q,m}=\sum_{p=0}^{N-1}\sum_{n=0}^{M-1} b_{p,n}\, s^{p,n}_{q,m}.$$3

and constant modulus implies

$$d_{q,m}=\sum_{p=0}^{N-1}\sum_{n=0}^{M-1} b_{p,n}\, s^{p,n}_{q,m}.$$4

The key analytical result is that any constant-modulus orthonormal basis yields identical OP performance once PIC produces the scalar per-symbol model. The effective scalar channel coefficient for symbol $$d_{q,m}=\sum_{p=0}^{N-1}\sum_{n=0}^{M-1} b_{p,n}\, s^{p,n}_{q,m}.$$5 is

$$d_{q,m}=\sum_{p=0}^{N-1}\sum_{n=0}^{M-1} b_{p,n}\, s^{p,n}_{q,m}.$$6

For constant-modulus sequences,

$$d_{q,m}=\sum_{p=0}^{N-1}\sum_{n=0}^{M-1} b_{p,n}\, s^{p,n}_{q,m}.$$7

which is identical for all $$d_{q,m}=\sum_{p=0}^{N-1}\sum_{n=0}^{M-1} b_{p,n}\, s^{p,n}_{q,m}.$$8 (Zemen et al., 2017). The theorem stated for the model therefore implies that DSFT, WHT, and any other constant-modulus orthogonal set yield identical per-symbol detection statistics and error probabilities under the derived scalar model.

This result directly addresses a frequent misunderstanding in discussions of OP for doubly selective channels: the gain is not tied uniquely to the DSFT basis. In the stated framework, the decisive property is constant modulus together with orthonormality, not the specific algebraic form of the basis (Zemen et al., 2017).

3. Iterative detection and the scalar decoupled model

The receiver is initialized at iteration $$d_{q,m}=\sum_{p=0}^{N-1}\sum_{n=0}^{M-1} b_{p,n}\, s^{p,n}_{q,m}.$$9 with an LMMSE window followed by matched filtering,

$\boldsymbol{d}=\boldsymbol{S}\,\boldsymbol{b},$0

Soft bits are produced by demapping and SISO decoding with BCJR, then re-interleaved and converted into soft symbol estimates $\boldsymbol{d}=\boldsymbol{S}\,\boldsymbol{b},$1 (Zemen et al., 2017).

For iterations $\boldsymbol{d}=\boldsymbol{S}\,\boldsymbol{b},$2, OPV2V-R uses PIC with a rectangular receive window $\boldsymbol{d}=\boldsymbol{S}\,\boldsymbol{b},$3. For each symbol index $\boldsymbol{d}=\boldsymbol{S}\,\boldsymbol{b},$4,

$\boldsymbol{d}=\boldsymbol{S}\,\boldsymbol{b},$5

This approximation yields the scalar effective channel model

$\boldsymbol{d}=\boldsymbol{S}\,\boldsymbol{b},$6

which decouples detection on a symbol-by-symbol basis (Zemen et al., 2017).

The associated symbol-wise ML decision rule is

$\boldsymbol{d}=\boldsymbol{S}\,\boldsymbol{b},$7

Soft-output metrics are then generated, for example by a soft-output sphere decoder, and passed to BCJR. The soft symbol update is

$\boldsymbol{d}=\boldsymbol{S}\,\boldsymbol{b},$8

With perfect PIC and accurate CSI, $\boldsymbol{d}=\boldsymbol{S}\,\boldsymbol{b},$9 has variance $\boldsymbol{S}$0, because $\boldsymbol{S}$1 under transmit orthonormality and unit-norm columns of $\boldsymbol{S}$2 (Zemen et al., 2017). Imperfect cancellation and CSI errors leave residual interference inside $\boldsymbol{S}$3, which inflates the effective noise variance; in the stated architecture, the BCJR decoder absorbs this through updated log-likelihood ratios.

The loop structure is fixed and explicit: initialization at $\boldsymbol{S}$4, followed by PIC, scalar detection, SISO decoding, and channel-estimation updates for $\boldsymbol{S}$5, with $\boldsymbol{S}$6 reported as empirically sufficient (Zemen et al., 2017). Final outputs are hard bits after BCJR at the last iteration, together with soft symbol decisions if HARQ is applicable.

4. Channel estimation with pilots and soft feedback

OPV2V-R incorporates iterative channel estimation based on pilots and soft symbols. If $\boldsymbol{S}$7 pilot positions and $\boldsymbol{S}$8 data positions satisfy $\boldsymbol{S}$9, and a permutation matrix $MN\times MN$0 multiplexes pilots $MN\times MN$1 and precoded data $MN\times MN$2, then the received signal can be written as

$MN\times MN$3

or equivalently

$MN\times MN$4

The iterative Wiener filter uses

$MN\times MN$5

together with a reliability matrix

$MN\times MN$6

where $MN\times MN$7 models soft-symbol uncertainty (Zemen et al., 2017). Using the channel covariance $MN\times MN$8, the update is

$MN\times MN$9

As the iterations proceed and $$\boldsymbol{y}=\operatorname{diag}(\boldsymbol{g})\,\boldsymbol{S}\,\boldsymbol{b}+\boldsymbol{n},$$0, the estimator reverts to classic Wiener filtering with pilots and reliable data decisions (Zemen et al., 2017). The formulation also permits reduced-rank implementation by exploiting the eigenstructure of $$\boldsymbol{y}=\operatorname{diag}(\boldsymbol{g})\,\boldsymbol{S}\,\boldsymbol{b}+\boldsymbol{n},$$1, which lowers matrix-inversion cost.

In the stated V2V setup, pilot density and interpolation are chosen to avoid aliasing in time and frequency. The receiver description further specifies $$\boldsymbol{y}=\operatorname{diag}(\boldsymbol{g})\,\boldsymbol{S}\,\boldsymbol{b}+\boldsymbol{n},$$2 dedicated pilot OFDM symbols at $$\boldsymbol{y}=\operatorname{diag}(\boldsymbol{g})\,\boldsymbol{S}\,\boldsymbol{b}+\boldsymbol{n},$$3, distributed across the frame as an improved pattern over IEEE 802.11p for non-line-of-sight channels (Zemen et al., 2017).

5. V2V operating regime and documented performance

The receiver is tailored to V2V links with relative velocity $$\boldsymbol{y}=\operatorname{diag}(\boldsymbol{g})\,\boldsymbol{S}\,\boldsymbol{b}+\boldsymbol{n},$$4 and is validated on IEEE 802.11p parameters (Zemen et al., 2017). The simulation setup uses $$\boldsymbol{y}=\operatorname{diag}(\boldsymbol{g})\,\boldsymbol{S}\,\boldsymbol{b}+\boldsymbol{n},$$5, bandwidth $$\boldsymbol{y}=\operatorname{diag}(\boldsymbol{g})\,\boldsymbol{S}\,\boldsymbol{b}+\boldsymbol{n},$$6, sample interval $$\boldsymbol{y}=\operatorname{diag}(\boldsymbol{g})\,\boldsymbol{S}\,\boldsymbol{b}+\boldsymbol{n},$$7, $$\boldsymbol{y}=\operatorname{diag}(\boldsymbol{g})\,\boldsymbol{S}\,\boldsymbol{b}+\boldsymbol{n},$$8 subcarriers, cyclic prefix length $$\boldsymbol{y}=\operatorname{diag}(\boldsymbol{g})\,\boldsymbol{S}\,\boldsymbol{b}+\boldsymbol{n},$$9 samples, and frame length $\boldsymbol{g}$0 OFDM symbols (Zemen et al., 2017). The channel model is geometry-based, non-stationary but WSS within a frame, with RMS delay spread $\boldsymbol{g}$1, a flat Doppler spectrum consistent with velocity, and representative urban V2V normalized supports $\boldsymbol{g}$2 up to $\boldsymbol{g}$3 and $\boldsymbol{g}$4 up to $\boldsymbol{g}$5. QPSK, a rate-$\boldsymbol{g}$6 convolutional code, and a random interleaver are assumed.

Under this setup, the reported BER results show that, compared with baseline OFDM without precoding, OP with DSFT and iterative PIC using three iterations achieves approximately $\boldsymbol{g}$7 SNR gain at $\boldsymbol{g}$8 for $\boldsymbol{g}$9 with perfect CSI (Zemen et al., 2017). Iterative gains are explicitly separated: iteration $\boldsymbol{n}$0 with LMMSE plus matched filtering provides partial diversity, iteration $\boldsymbol{n}$1 with PIC and symbol-wise ML adds approximately $\boldsymbol{n}$2, and iteration $\boldsymbol{n}$3 adds approximately $\boldsymbol{n}$4; further iterations yield no practical improvement.

With pilot-based channel estimation and iterative Wiener updates, the performance curves shift by approximately $\boldsymbol{n}$5 because of CSI error, while the qualitative OP gains remain (Zemen et al., 2017). The reported basis comparison is equally specific: DSFT and WHT are identical in performance, whereas 2D-DPS can be approximately $\boldsymbol{n}$6 better at $\boldsymbol{n}$7, which the source associates with slightly improved matching to channel correlation.

The performance interpretation given for OPV2V-R rests on three coupled mechanisms. First, OP averages the channel across the entire frame, producing channel hardening in the time–frequency domain:

$\boldsymbol{n}$8

Second, PIC removes the multi-symbol coupling introduced by the doubly selective channel acting on the precoded frame. Third, SISO decoding benefits from more stable soft information (Zemen et al., 2017). This suggests that the performance advantage is not merely a coding gain or a transform gain in isolation; it arises from the interaction of spreading, iterative interference cancellation, and iterative decoding.

6. Complexity, compatibility, and limitations

The computational profile of OPV2V-R is dominated by the precoder or its inverse transform, PIC projections, channel estimation, and BCJR decoding. General OP has complexity $\boldsymbol{n}$9, but the DSFT can be implemented with two FFTs at complexity $n_{q,m}\sim\mathcal{CN}(0,\sigma_n^2)$0 (Zemen et al., 2017). With rectangular windows, the DSFT FFT and OFDM IFFT can be combined, leaving $n_{q,m}\sim\mathcal{CN}(0,\sigma_n^2)$1 at both transmitter and receiver. For DSFT and WHT, the PIC projections are fast-transform operations, so per-iteration cost is also $n_{q,m}\sim\mathcal{CN}(0,\sigma_n^2)$2. The source further notes that soft-output per-symbol ML for QPSK is low complexity, because sphere decoding reduces to closed-form metrics.

The architecture is presented as standards-compatible in the sense that the OP layer sits above OFDM and remains compatible with IEEE 802.11p PHY framing, while requiring modified pilot placement and receiver processing but no change to the RF front-end or OFDM parameters (Zemen et al., 2017). Recommended parameters are explicitly listed as $n_{q,m}\sim\mathcal{CN}(0,\sigma_n^2)$3, $n_{q,m}\sim\mathcal{CN}(0,\sigma_n^2)$4, $n_{q,m}\sim\mathcal{CN}(0,\sigma_n^2)$5, $n_{q,m}\sim\mathcal{CN}(0,\sigma_n^2)$6, $n_{q,m}\sim\mathcal{CN}(0,\sigma_n^2)$7, QPSK, rate-$n_{q,m}\sim\mathcal{CN}(0,\sigma_n^2)$8 convolutional coding, random interleaving, DSFT or WHT basis, rectangular windows, two to three iterations, and $n_{q,m}\sim\mathcal{CN}(0,\sigma_n^2)$9 pilot symbols per frame.

Several implementation conditions are also stated. Accurate time and frequency synchronization is necessary to preserve the diagonalized OFDM model; residual CFO and timing errors increase ICI and degrade PIC (Zemen et al., 2017). Doppler tracking is delegated to the iterative Wiener estimator using $b_{p,n}$00, and reduced-rank subspace tracking based on known delay and Doppler supports is proposed for stability and lower complexity. The documented limitations are equally concrete: extremely severe Doppler beyond the assumed supports, or insufficient CP for long delays, breaks the diagonal model through increased ICI or ISI and thereby reduces PIC effectiveness. Pilot overhead and iterative decoding also raise computational load, so embedded implementations require careful parameter selection.

An additional analytical metric used to characterize hardening is

$b_{p,n}$01

where $b_{p,n}$02 are the eigenvalues of $b_{p,n}$03 (Zemen et al., 2017). Larger delay and Doppler supports reduce $b_{p,n}$04, which the source identifies as the reason gains increase in more strongly doubly selective channels.

7. Significance within doubly selective V2V reception

Within the documented framework, OPV2V-R is notable because it provides a fully documented receiver architecture for coded OP in doubly selective channels and reports that its coded OP performance is the best among fully documented receiver architectures published at that time (Zemen et al., 2017). Its significance lies in converting a heavily coupled two-dimensional time–frequency detection problem into a scalar per-symbol problem through PIC, while retaining the diversity benefit of frame-wide spreading.

The architecture also clarifies the role of basis choice in OP-based V2V reception. DSFT is not privileged by performance under the constant-modulus theorem; WHT is interchangeable in that respect, while 2D-DPS offers only a marginal gain at higher complexity (Zemen et al., 2017). This makes basis selection primarily a question of implementation efficiency and channel matching rather than of a fundamentally different detection principle.

From a system perspective, the reported throughput and latency implications are limited but explicit. For fixed code rate and modulation, the OP SNR gain can be traded for higher throughput via modulation or coding upgrades, or for improved reliability via lower BER, while latency remains dominated by frame duration and a small number of decoding iterations no greater than three (Zemen et al., 2017). A plausible implication is that the receiver was formulated not as a generic transform-domain enhancement, but as a design targeted at ultra-reliable wireless communication in high-mobility V2V environments where both delay and Doppler diversity are substantial.