CTRF: Coordinated Transmit-Receive Flexibility

• CTRF is a coordinated transceiver design principle where transmit and receive operations are jointly optimized to flexibly adapt streams and mitigate coupling effects.
• It enables adaptive stream selection in MU-MIMO and OFDMA systems by dynamically varying stream counts to improve SNR, reduce BER, and maximize sum-rate.
• CTRF extends to MRI and sub-Nyquist delay-Doppler estimation, using active analog cancellation and joint filter optimization to suppress interference and enhance resolution.

Searching arXiv for the cited CTRF-related papers and nearby work. arxiv_search query: Coordinated-Transmit-Receive-Flexible CTRF beamforming OR coordinated transmit and receive processing with adaptive multi-stream selection OR Active Decoupling of Transmit and Receive Coils for Full-Duplex MRI OR Joint Transmit and Receive Filter Optimization for Sub-Nyquist Delay-Doppler Estimation

Coordinated-Transmit-Receive-Flexible (CTRF) denotes, in the cited literature, transceiver designs in which transmit-side and receive-side operations are jointly coordinated while the active degrees of freedom are flexibly adapted to channel, stream, or coupling conditions. The term is used most explicitly for flexible stream allocation in MU-MIMO beamforming and OFDMA uplink zero-forcing, and it also appears in CTRF MRI, where concurrent excitation and acquisition are enabled by active cancellation of transmit leakage into the receive chain. A related line of work studies joint transmit/receive filter optimization for sub-Nyquist delay-Doppler estimation under Bayesian criteria (An et al., 2010, Marques et al., 28 Sep 2025, Salim et al., 2018, Lenz et al., 2017).

1. Terminology and research scope

Across the available literature, CTRF is not confined to a single implementation. It appears in at least three technically distinct but structurally related settings: adaptive coordinated Tx–Rx beamforming for multi-user downlink transmission, flexible zero-forcing stream allocation for OFDMA uplink reception, and concurrent excitation/acquisition MRI with active analog cancellation. A related coordinated transceiver-design formulation optimizes analog transmit and receive filters jointly for delay-Doppler estimation.

Domain	Flexible coordinated element	Representative source
Narrow-band MU-MIMO downlink	Adaptive per-user stream count $L_k$ with coordinated Tx–Rx beamforming and block diagonalization	(An et al., 2010)
OFDMA MU-MIMO uplink	Flexible per-PRB stream allocation with ZF reception and TS-wide RRM	(Marques et al., 28 Sep 2025)
Concurrent excitation and acquisition MRI	Analog cancellation path using attenuation and delay matching of Tx$\rightarrow$Rx coupling	(Salim et al., 2018)
Delay-Doppler estimation	Joint analog transmit/receive filter optimization under BCRLB with $B>f_s$	(Lenz et al., 2017)

This spread of usage suggests that CTRF is best understood as a coordinated transceiver principle rather than a single fixed algorithm. The common motif is that transmit and receive processing are not optimized independently: the transmitter, receiver, and stream or filter selection are co-designed against a performance criterion such as BER, weighted sum-rate, isolation, or Bayesian estimation accuracy.

2. Downlink origin: adaptive coordinated Tx–Rx beamforming

A foundational formulation appears in the narrow-band MU-MIMO downlink model where a single base station with $N_t$ antennas serves $K$ users, user $k$ has $N_{r,k}$ receive antennas, and the transmitted signal is

$x=\sum_{k=1}^K T_k D_k,$

with $T_k \in \mathbb{C}^{N_t \times L_k}$ and a total power constraint $\sum_k \mathrm{Trace}(T_kT_k^H)\le P_T$. User $\rightarrow$0 observes

$\rightarrow$1

and after receive-side preprocessing $\rightarrow$2 obtains

$\rightarrow$3

In the conventional coordinated Tx–Rx beamforming scheme, the number of streams is fixed a priori as $\rightarrow$4 for all users, irrespective of instantaneous channel quality (An et al., 2010).

The fixed-stream design proceeds in two stages. First, each user forms a pre-receiver filter $\rightarrow$5 and an equivalent channel block $\rightarrow$6. Second, a block-diagonalizing transmit design enforces $\rightarrow$7 for all $\rightarrow$8, typically through two SVD/QR passes: the right singular vectors associated with the zero-singular-value subspace of

$\rightarrow$9

define $B>f_s$0, then $B>f_s$1 projected onto $B>f_s$2 is decomposed again and the top-$B>f_s$3 right singular vectors $B>f_s$4 are selected, giving $B>f_s$5. After this transmit design, each user recomputes $B>f_s$6 to maximize post-processing SNR.

The central limitation of the fixed-stream formulation is that ill-conditioned channels suffer severe noise amplification in zero-forcing or block diagonalization. The adaptive CTRF scheme addresses this by allowing the stream count to vary per user, $B>f_s$7, according to instantaneous singular-value structure. Using the SVD

$B>f_s$8

with singular values $B>f_s$9, the post-equalization noise amplification of the $N_t$0-th mode in a ZF-type receiver is proportional to $N_t$1. This motivates either

$N_t$2

or a sum-rate proxy

$N_t$3

under the same constraints.

A practical sub-optimal stream-selection method uses sorted QR-decomposition on $N_t$4 and a cumulative metric

$N_t$5

Streams are added greedily in descending order of strength, while fairness is enforced by guaranteeing each user at least one stream. Once the vector $N_t$6 is chosen, the coordinated Tx–Rx design is re-run with user-dependent stream counts. Under simulation with total streams $N_t$7, four users, $N_t$8, $N_t$9 each, and either ZF or MMSE detection, the adaptive CTRF achieves approximately a $K$0 SNR saving at $K$1 compared to the conventional fixed-2-streams-per-user scheme (An et al., 2010).

3. Uplink CTRF as a zero-forcing stream-allocation strategy

A later and more explicit definition of Coordinated-Transmit-Receive-Flexible appears in the uplink of a single-cell OFDMA MU-MIMO system. The cell is divided into $K$2 OFDMA subchannels per time-slot, there are $K$3 users, user $K$4 has $K$5 transmit antennas, and the base station has $K$6 receive antennas. On PRB $K$7, the received vector is

$K$8

where $K$9 and $k$0. Each user has a TS-wide power budget

$k$1

In CTRF, each user may send up to $k$2 streams per PRB, but the scheduler flexibly chooses which subset is active (Marques et al., 28 Sep 2025).

The selection variable $k$3 indicates whether stream $k$4 of user $k$5 is active on PRB $k$6, with the ZF stream-count constraint

$k$7

All active user-stream channel columns are stacked into

$k$8

where $k$9 and $N_{r,k}$0 is the $N_{r,k}$1-th right-singular vector of $N_{r,k}$2 or the $N_{r,k}$3-th column of $N_{r,k}$4 after a suitable decomposition. The ZF receive filter is the Moore–Penrose pseudo-inverse

$N_{r,k}$5

so that $N_{r,k}$6 and inter-stream interference is nulled.

For stream $N_{r,k}$7 on PRB $N_{r,k}$8, the post-combining signal is

$N_{r,k}$9

and, because of ZF nulling, the remaining degradation apart from noise is beam-forming loss. The effective channel gain is

$x=\sum_{k=1}^K T_k D_k,$0

and the SINR with allocated power $x=\sum_{k=1}^K T_k D_k,$1 is

$x=\sum_{k=1}^K T_k D_k,$2

A practical MCS-based rate function $x=\sum_{k=1}^K T_k D_k,$3 maps SINR to spectral efficiency, so the total rate of user $x=\sum_{k=1}^K T_k D_k,$4 in the time-slot is

$x=\sum_{k=1}^K T_k D_k,$5

This formulation makes the “flexible” component of CTRF explicit. Unlike CTR1, where only the strongest data stream is enabled per scheduled user, or BD, where all possible streams are enabled per scheduled user, CTRF allows a flexible stream allocation per user. The additional flexibility expands the feasible ZF operating set, but also makes scheduling and power allocation intrinsically coupled.

4. Time-slot-wide optimization, heuristics, and complexity

The uplink CTRF scheduler solves a TS-wide radio-resource-management problem that maximizes a weighted sum-rate

$x=\sum_{k=1}^K T_k D_k,$6

subject to the ZF stream-count constraint, the TS-power constraint, the per-user stream cap, $x=\sum_{k=1}^K T_k D_k,$7, and $x=\sum_{k=1}^K T_k D_k,$8. Because $x=\sum_{k=1}^K T_k D_k,$9 is nonlinear, $T_k \in \mathbb{C}^{N_t \times L_k}$0 is discrete, and the ZF gains $T_k \in \mathbb{C}^{N_t \times L_k}$1 depend on the selected stream set, the problem is a mixed-integer, non-convex program and cannot be solved exactly for realistic system sizes (Marques et al., 28 Sep 2025).

To obtain feasible real-time solutions, the cited work proposes a Greedy-Up Search (GUS). The procedure initializes all $T_k \in \mathbb{C}^{N_t \times L_k}$2 to zero and iteratively tests inactive candidate streams. For each candidate $T_k \in \mathbb{C}^{N_t \times L_k}$3, the algorithm temporarily activates the stream, recomputes ZF effective gains on the affected PRB using fast update formulas, solves the relevant single-user power-management subproblems through fitted-MCS, concave log-utility, and water-filling, evaluates the tentative weighted sum-rate, and then undoes the provisional choice. The candidate producing the largest gain is retained if it improves the incumbent weighted sum-rate by a factor larger than $T_k \in \mathbb{C}^{N_t \times L_k}$4 with $T_k \in \mathbb{C}^{N_t \times L_k}$5; otherwise the search stops. A final pass performs exact MCS-aware power-management through greedy MCS-upgrade with leftover power, prunes zero-rate streams, recomputes ZF on affected PRBs, and redoes power-management.

The heuristic’s complexity is characterized in two complementary ways. At the per-iteration level, each iteration examines $T_k \in \mathbb{C}^{N_t \times L_k}$6 candidates, and each candidate requires one ZF update plus $T_k \in \mathbb{C}^{N_t \times L_k}$7 water-filling solves. At the asymptotic level, the overall complexity is roughly

$T_k \in \mathbb{C}^{N_t \times L_k}$8

where $T_k \in \mathbb{C}^{N_t \times L_k}$9 is the number of search iterations and typically $\sum_k \mathrm{Trace}(T_kT_k^H)\le P_T$0. This establishes the main engineering trade-off of CTRF in the uplink setting: additional degrees of freedom improve feasibility and rate, but only at the cost of a more difficult combinatorial selection problem.

5. Comparative performance, fairness, and operating regimes

The performance analysis of uplink ZF strategies compares CTRF against CTR1 and BD using average sum-rate, geometric-mean throughput, Jain’s Fairness Index, and computational complexity. The extra degrees of freedom in CTRF strictly improve or match BD and CTR1, but at higher complexity. The numerical outcomes, however, show that the relative advantage depends strongly on propagation scenario, user density, antenna dimensions, and power-management fidelity (Marques et al., 28 Sep 2025).

In Rural Macro scenarios with $\sum_k \mathrm{Trace}(T_kT_k^H)\le P_T$1, the reported behavior is regime-dependent: for $\sum_k \mathrm{Trace}(T_kT_k^H)\le P_T$2, BD is within $\sum_k \mathrm{Trace}(T_kT_k^H)\le P_T$3 of CTRF, whereas for $\sum_k \mathrm{Trace}(T_kT_k^H)\le P_T$4, CTR1 is within $\sum_k \mathrm{Trace}(T_kT_k^H)\le P_T$5 of CTRF. In Urban Macro scenarios, CTR1 trails CTRF by less than $\sum_k \mathrm{Trace}(T_kT_k^H)\le P_T$6 for all $\sum_k \mathrm{Trace}(T_kT_k^H)\le P_T$7, while BD lags by $\sum_k \mathrm{Trace}(T_kT_k^H)\le P_T$8–$\sum_k \mathrm{Trace}(T_kT_k^H)\le P_T$9 except at very low $\rightarrow$00. Raising $\rightarrow$01 from $\rightarrow$02 to $\rightarrow$03 yields $\rightarrow$04–$\rightarrow$05 rate gains for all strategies. Increasing $\rightarrow$06 in UMa from $\rightarrow$07 to $\rightarrow$08 roughly doubles rates; the CTRF-to-CTR1 gap widens slightly, whereas the BD gap remains stable. A simple equal-power-per-stream scheme costs $\rightarrow$09 for CTRF and CTR1 but up to $\rightarrow$10 for BD relative to Two-step PM. The rate-reusing speedup reduces heuristic runtime by about $\rightarrow$11 for CTR1 and $\rightarrow$12 for BD without performance loss.

These findings delimit where CTRF’s flexibility is most valuable. If real-time complexity is acceptable, CTRF should be used to harvest all spatial-multiplexing gains. If complexity is constrained, the data indicate that BD can be near-optimal in RMa with moderate user counts, and CTR1 can be sufficient for very large $\rightarrow$13 in RMa or across UMa deployments. The result is not that CTRF is uniformly dominant in practical deployment, but that it defines the upper envelope against which lower-complexity ZF strategies are measured.

6. CTRF MRI and related coordinated transceiver co-design

In MRI, CTRF refers to a concurrent excitation and acquisition architecture whose central difficulty is the enormous disparity between transmit and receive amplitudes: transmit power is typically tens of watts, whereas the received MR signal voltage is on the order of microvolts. Even weak Tx-to-Rx coupling therefore overwhelms the MR signal. The active decoupling strategy samples a fraction of the transmit RF waveform, imposes the same attenuation and delay introduced by the Tx$\rightarrow$14Rx coupling path, and subtracts this anti-signal from the raw receiver output. If done ideally, the transmit leakage is canceled down to the receiver’s noise floor, leaving only the true MR signal (Salim et al., 2018).

The coupling model is written as

$\rightarrow$15

where $\rightarrow$16 is the electrical transfer function from the Tx port to the Rx port. The injected cancellation waveform is

$\rightarrow$17

so the residual coupling becomes

$\rightarrow$18

At a center frequency $\rightarrow$19, if

$\rightarrow$20

the zero-residual condition is

$\rightarrow$21

More generally, one may minimize the mean-square residual over a small band by optimizing $\rightarrow$22 and $\rightarrow$23.

Two controllable decoupling designs are described. The Semi-Automatic Controllable Decoupling Design uses an unequal $\rightarrow$24 Wilkinson divider, a fixed coaxial delay line, a $\rightarrow$25-step programmable attenuator (HMC759), and a Wilkinson combiner; calibration is performed with an Agilent E5061B network analyzer at $\rightarrow$26 while software sweeps the attenuator in $\rightarrow$27 increments. The Fully-Automatic Controllable Decoupling Design uses four parallel cancellation branches with fixed $\rightarrow$28, $\rightarrow$29, $\rightarrow$30, and $\rightarrow$31 phase shifters, four identical HMC759 attenuator ICs, branch-selection logic that activates exactly two branches whose phase axes enclose the coupling phase $\rightarrow$32, and a Raspberry Pi running a genetic-algorithm optimizer. Measurements at the $\rightarrow$33 Larmor frequency with orthogonal coils providing about $\rightarrow$34 passive geometric decoupling showed peak decoupling above $\rightarrow$35 and a $\rightarrow$36 cancellation bandwidth of about $\rightarrow$37 for the semi-automatic design, and peak decoupling above $\rightarrow$38 with about $\rightarrow$39 bandwidth where at least $\rightarrow$40 cancellation is maintained for the fully automatic design.

For integration into a full CTRF MRI platform, the practical considerations are explicit: insert the unequal Wilkinson divider at the RF transmit port of each channel; route the cancellation tap through programmable attenuators and a phase-shift network; recombine ahead of the low-noise amplifier; phase-match cable lengths and PCB traces to within a few degrees across the band; run fast optimization between pulse sequences; optionally close the loop on raw digitized Rx data by computing the sample covariance of the first few microseconds of each acquisition; maintain stability under temperature and loading changes, including patient motion and coil detuning; preserve more than $\rightarrow$41 margin to the noise floor under worst-case detuning; and trigger shutdown if residual power above $\rightarrow$42 reaches the receiver front end. In this setting, CTRF extends coordinated transmit/receive processing from spatial stream management to full-duplex analog leakage suppression.

A related but distinct coordinated transceiver-design problem appears in sub-Nyquist delay-Doppler estimation, where the analog transmit pulse $\rightarrow$43 and receive filter $\rightarrow$44 are jointly optimized under a Bayesian Cramér–Rao lower bound with $\rightarrow$45. The model uses a periodic transmit waveform

$\rightarrow$46

a single-path channel

$\rightarrow$47

and samples

$\rightarrow$48

The design minimizes weighted MSE through the BCRLB, equivalently maximizing an approximation to $\rightarrow$49 under a transmit-power constraint on $\rightarrow$50. The resulting alternating optimization updates $\rightarrow$51 as the principal eigenvector of a Hermitian quadratic form $\rightarrow$52 and each receive-filter frequency-bin component as the principal eigenvector of $\rightarrow$53. For $\rightarrow$54, $\rightarrow$55, $\rightarrow$56, $\rightarrow$57, and $\rightarrow$58, the optimized design yields about $\rightarrow$59 gain in $\rightarrow$60-MSE and about $\rightarrow$61 gain in $\rightarrow$62-MSE over a Nyquist-limited rectangular-pulse reference in the high-SNR region, with break-even SNRs around $\rightarrow$63-Hz for delay and $\rightarrow$64-Hz for Doppler (Lenz et al., 2017).

Taken together, these strands show that CTRF is best read as a coordinated design doctrine with multiple realizations. In MU-MIMO it denotes flexible stream budgeting under joint transmit/receive processing; in OFDMA uplink it becomes a TS-wide ZF scheduling and power-allocation strategy; in MRI it denotes full-duplex concurrent excitation and acquisition through controllable analog cancellation; and in related estimation problems it motivates joint analog transmit/receive filter optimization under explicit statistical criteria.