Software-Defined Radio Experimental Validation of an OTFS-Based ISAC for Velocity Estimation in an ARoF Setup

Abstract

OTFS, proposed for next-generation wireless communication systems such as 6G mobile networks, incorporates ISAC into DD-domain multiplexing, enabling simple detection of distance, velocity, and movement direction. This paper presents a SDR implementation of OTFS in an ARoF setup with wireless RF transmission. The main goal of this study is to validate and evaluate the implemented OTFS with static objects and to explore the quality of velocity and direction estimation in sensing scenarios involving moving objects. For the BER measurements, experiments were performed using a static object while varying the SDR transmitter power and introducing additional CFO. Experimental validation shows a minimum BER ≤ 5 × 10−⁷ with 0 errors per 2 × 10⁶ bits. Data transmission at fractional Doppler yielded a BER ≈ 0.09, which is attributed to the use of a LMMSE channel estimator, that is not optimal for channels with fractional Doppler. Estimation of the velocity of a mobile object with an absolute velocity of $\left|v\right|=0.15$ m/s yielded a RMSE = 0.0839 m/s.

1. Introduction

With the increased need for stable communications with and between high-mobility UEs, such as V2V, V2X, LEO satellites, and highly mobile IoT devices, waveforms and modulation systems capable of demodulating data in the presence of Doppler shift are required [1]. Existing implementations of OFDM exhibit performance deterioration in high-mobility scenarios [2,3,4,5]. To address this issue, different OFDM-based systems have been proposed, simulated, and tested [6,7,8], with the main goal of improving communication performance in high-mobility scenarios. Doppler-assisted channel estimation for high-mobility communication using classical OFDM was also explored in many works [9,10].

One of the multicarrier modulation formats designed for communication with high-mobility UEs, first published by R.Hadani et al. [11] in 2017, is OTFS. This format was initially based on OFDM, with an additional pre/post-modulator that employs Heisenberg and Wigner transforms to achieve frequency-domain orthogonality. With the introduction of the DZT into the signal processing pipeline [12,13,14], the implementation of OTFS became simpler. IDZT directly converts DD-domain symbols into the time-domain transmit signal, while the corresponding inverse operation at the receiver directly recovers the DD-domain symbols, thereby replacing the separate IFFT-based/Heisenberg modulation and Wigner/FFT-based demodulation stages with a unified transform pair. OTFS is also being proposed as part of 6G [15,16] with NTN integration in mind. In OTFS, data are multiplexed in the DD domain, which facilitates the estimation of the delay and Doppler of received signal reflections, and thus enables the detection of target distance and velocity. The OTFS DD domain matrix is shown in Figure 1, where reflections from static and mobile objects are present at the receiving base station, which performs the sensing function. These marked DD cells have their own delay and Doppler tap values, which correspond to the reflected pilot pulses transmitted by the sensing base station.

To reduce costs associated with the introduction of next-generation communication systems, e.g., 6G, existing base stations with RAUs can be connected to the CU, located tens of kilometers away. A CRAN scenario where the optical link is employed to connect the CU to base and satellite stations is depicted in Figure 1.

As the generated RF signal must be transmitted over multiple kilometers through the fiber, ARoF-capable waveforms should be employed. Existing research shows that OTFS can be transmitted using an ARoF setup [17]. Special types of OTFS [18,19,20] are intended to deal with OWC environments, such as VLC, where the envelope of the RF signal should be real and positive due to the employment of intensity-modulation and direct detection.

SDR enables fast prototyping of radio communication systems and is a significantly cheaper alternative to arbitrary waveform generators. Multiple practical implementations of OTFS systems on SDR exist [20,21,22], with the first implementation conducted by T.Thaj and E.Vitebro in 2019 [23]. OTFS ISAC implementations on SDRs also exist [24,25,26,27,28], and the number of reported successful implementations is increasing each year.

Most of the mentioned implementations use direct wireless analog transmission from the SDR [21,22,23,24,25,26,27,28] or employ arbitrary waveform generators and digital oscilloscopes to generate and receive OTFS waveforms [17,29,30]. In this paper, SDR hardware is employed in the prototyping of a real-time communication system in ARoF with wireless RF communication. OTFS system on SDR is implemented in the ARoF setup with an 800 m long SMF span. This paper provides a full description of the optical setup in use, which is comprised of the minimum number of hardware components required to achieve an ARoF link between the SDR’s input and output ports. In comparison with other implementations of OTFS, which are primarily meant for V2X or vehicle-to-base station scenarios for high-mobility communications [26,27,31], this study addresses very low velocity < 1 km/h detection of a moving object at the cost of distance resolution. The sensing part of this OTFS ISAC system estimates the velocity of an object that reflects the RF OTFS signal in real time, essentially performing the role of a monostatic radar for object velocity and movement direction detection. It is worth noting that the use of ARoF not only increases the maximum allowed distance between the antennas and the CU, but can also enable distributed optical fiber sensing capabilities [32] in addition to the implemented ISAC.

This work is divided into the following sections: Section 2 describes an SDR implementation of an OTFS system, while Section 3 provides the description of an optical setup used in the experiments. Results from the conducted experiments are available in Section 4, with discussion and conclusions in Section 5 and Section 6, respectively.

2. Software-Defined Radio Implementation of an OTFS System

A real-time OTFS communication system for ISAC validation was implemented in MATLAB Simulink R2025a running on a PC connected to the USRP B210 (Ettus Research, Austin, TX, USA) SDR. The block diagram of the implemented OTFS transceiver Simulink model is presented in Figure 2.

First, at the transmitter side, QAM data is encoded in the DD domain—$M\times N$ matrix, where N is the number of time slots (Doppler bins) and M is the number of subcarriers (delay bins). Then IDZT is applied to transform the DD domain into the time domain for transmission. IDZT can be split into two major steps: application of IFFT and P/S conversion. First, QAM data $\mathit{X}$ in the DD domain is moved to the DT domain via the application of IFFT along Doppler dimension:

$$Y\left[m,n\right]={\displaystyle \frac{1}{N}}\sum _{l=0}^{N-1}X\left[k,l\right]e^{2\pi j\left({\displaystyle \frac{n·l}{N}}\right)} ; k\in [0,M-1] ,$$

(1)

where $\mathit{X}$ is an $M\times N$ DD domain matrix with QAM encoded data, $\mathit{Y}$ is a DT domain matrix, before P/S conversion into the time domain, which is necessary for transmission into a real-world communication channel. After the IFFT step, column-by-column, or in this case, column is a sub-carrier, P/S conversion is applied to get a time domain signal vector $\mathit{s}$ using $\mathit{Y}$ from Formula (1):

$$s\left[k\right]=Y\left[k \mathrm{mod} M,⌊{\displaystyle \frac{k}{M}}⌋\right] ; k\in [0,N·M-1] .$$

(2)

The positions of data with ZP and the pilot in the DD and DT domains before and after the IFFT, applied by (1), are shown in Figure 3.

After the P/S conversion, a CP is added to enable synchronization at the receiver side:

$$s_{\mathrm{CP}}\left[k\right]=s\left[(k-M_{\mathrm{CP}}) \mathrm{mod} (N·M)\right] ; k\in [0,N·M+M_{\mathrm{CP}}-1] ,$$

(3)

where $M_{\mathrm{CP}}$ is a CP length in samples, and ${\mathit{s}}_{\mathrm{CP}}$ is a time domain signal vector with appended CP in the front of the signal $\mathit{s}$.

Before the transmission, up-sampling of a discrete signal ${\mathit{s}}_{\mathrm{CP}}$ is done by the repetition of each sample $r=10$ times:

$$s_{\mathrm{up}}\left[k\right]=s_{\mathrm{CP}}\left[⌊{\displaystyle \frac{k}{r}}⌋\right] ; k\in [0,r·(N·M+M_{\mathrm{CP}})-1] ,$$

(4)

where ${\mathit{s}}_{\mathrm{up}}$ is an up-sampled discrete signal vector that will be sent to the SDR for transmission.

On the receiver side, the received signal is split between the CP correlator and the fractional delay blocks. CP correlation is needed to estimate the fractional delay of the received signal, so that the OTFS frame is passed to the rest of the system at the right starting sample. If the starting sample is delayed by a value outside the channel estimation window, demodulation will yield errors in the received data. In the radar system, this sudden sample delay will show as a non-zero distance if the transmitted signal is synchronized to the hardware propagation time delay.

After synchronization via CP correlation, the CP (the first $M_{\mathrm{CP}}$ samples) is removed from the time-domain signal, and down-sampling by an r order is performed. The received signal will be altered by delay and Doppler effects, which can be expressed by the conventional multi-path Doppler channel model:

$$\widehat{s}\left[n\right]=\sum _{i=1}^K{h}_i{e}^{j2\pi {\nu }_{i}n}s\left[{\left(n-{\tau }_{i}B\right)}_{MN}\right] , n=0,\dots ,MN-1 ,$$

(5)

where K is the number of channel paths, each characterized by a complex gain $h_i$, delay ${\tau }_i$, and Doppler shift ${\nu }_i$. B denotes the system bandwidth, ${\left(·\right)}_{MN}$ is the modulo-$MN$ operator, and $\widehat{s}$ is a serial discrete signal of size $M·N$.

Both signals, before and after the first DZT, are used in the LMMSE channel estimation. DZT, similarly to the IDZT used in the transmitter, can be split into S/P and FFT parts, expressed by Formulas (6) and (7), respectively:

$$\widehat{Y}[m,n]=\widehat{s}[n·M+m] ; n\in [0,N-1] ; m\in [0,M-1] ,$$

(6)

where $\widehat{\mathit{Y}}$ is a DT domain matrix of size $M\times N$ after S/P and before FFT along the time dimension:

$$\widehat{X}[m,n]=\sum _{l=0}^{N-1}\widehat{Y}[k,l]e^{-2\pi j\left({\displaystyle \frac{n·l}{N}}\right)} ; k\in [0,M-1] ,$$

(7)

where $\widehat{\mathit{X}}$ is a DD domain matrix after application of FFT in DZT.

Before LMMSE is performed, the channel in the DD domain from the first DZT is normalized and then used for object velocity estimation. The equalized signal from the LMMSE estimator is then converted to the DD domain via a second DZT using the previously defined Formulas (6) and (7). After DZT, QAM data are decoded, and the resulting binary data are used to calculate BER, which is used in the system’s performance evaluation.

Parameters of the implemented OTFS communication system on the USRP B210 SDR are shown in

Table 1. Parameters of the SDR OTFS ISAC system.

Parameter	Description	Value
$f_{\mathrm{c}}$	Carrier frequency, GHz	6
$\lambda =\mathrm{c}/(2·f_{\mathrm{c}})$	Carrier wavelength, m	$0.025$
N	Number of time slots (Doppler bins)	10
M	Number of subcarriers (delay bins)	16
$M_{\mathrm{CP}}$	Length of the CP in samples	16
$M_{\mathrm{ZP}}$	Length of the ZP in delay taps	4
T	Frame time, ms	7
$T_{\mathrm{r}}=T/(M·N+M_{\mathrm{CP}})$	Single tap delay resolution, μs	40
$d_{\mathrm{r}}=T_{\mathrm{r}}·\mathrm{c}/2$	Single tap distance resolution, km	$5.996$
$F_{\mathrm{SDR}}$	SDR bandwidth, kHz	250
F	OTFS signal bandwidth, kHz	$22.727$
$F_{\mathrm{r}}=F/(M·N)$	Single tap Doppler resolution, Hz	$142.046$
$[-(F_{\mathrm{r}}·N)/2,(F_{\mathrm{r}}·N)/2-F_{\mathrm{r}}]$	Doppler range, Hz	$[-710.227,+568.182]$
$v_{\mathrm{r}}=F_{\mathrm{r}}·\lambda$	Single tap velocity resolution, m/s	$3.549$

.

The OTFS signal bandwidth F and numbers of bins were chosen specifically to achieve velocity estimation resolution $v_{\mathrm{r}}=F_{\mathrm{r}}/f_{\mathrm{c}}·c/2=3.549 \mathrm{m}/\mathrm{s}$, because movement of a large metal plate used as a signal reflector at such a velocity can be performed under laboratory conditions. Given the limitations of the experimental setup and OTFS signal parameters, the distance cannot be measured without special delays, as a minimum distance of $d_{\mathrm{r}}=T_{\mathrm{r}}·\mathrm{c}/2=5.996 \mathrm{k}\mathrm{m}$ to the object is required to measure the single-tap delay of the pilot signal.

3. Experimental Setup

The experimental setup consists of the CU and the RAU, connected by 400 m long SMF spans. The OTFS waveform was generated by the USRP B210 SDR using parameters from Table 1. The “TX” port of the SDR, with a maximum power of $-2.4$ $\mathrm{dBm}$, was connected to the MX-LN-40 MZM (Exail Technologies, Paris, France), which features a 40 GHz bandwidth, a 20 dB extinction ratio, and a 3.5 dB insertion loss. The optical carrier was generated by the DX-4 CW laser (ID Photonics, Neubiberg, Germany), which has a line width of 25 kHz. The wavelength of the optical carrier was 1550 nm. The optical power of the carrier +2.4 dBm was split using a 50/50 optical power splitter and sent through the polarization controllers.

The optical output of the MZM was connected to the KG-PT-10G-SM-FA InGaAs/InP planar structure PIN PD (Conquer, Beijing, China) through a 90/10 optical power splitter for remote power monitoring purposes. The PD features a 10 GHz bandwidth, a dark current of 10 nA, and a responsivity of $0.9$ $\mathrm{A}/\mathrm{W}$. The OPM was used to maintain an optical power of $-15$ $\mathrm{dBm}$ on the PD, as the PD’s saturation limit is $-11$ $\mathrm{dBm}$. The PD’s RF output was connected to the SHF 810 broadband RF amplifier (SHF Communication Technologies AG, Berlin, Germany) with 29 $\mathrm{d}\mathrm{B}$ of gain and a 38 GHz bandwidth. The amplified signal was fed into the WR-159 horn antenna (Pasternack, Irvine, CA, USA), which has a gain of 20 $\mathrm{dBi}$ and a frequency range of 4.9–7.05 GHz.

The same antenna model was used at the receiving side. A SHF 100 BP broadband amplifier with 17 $\mathrm{d}\mathrm{B}$ gain and 25 GHz bandwidth was used to amplify the received signal. The amplifier’s output was connected to the MZM in the RAU, and its optical output was passed to the CU. The optical signal was split for power monitoring and passed to the PD located in the CU. The PD’s output was connected to the SDR’s “RX” port.

In the present 800-m standard-SMF ARoF setup at a 6 GHz RF carrier, chromatic dispersion does not introduce appreciable RF power fading or noticeable spreading of the recovered OTFS DD representation. Because the link is based on direct detection, laser phase noise is not expected to be a dominant source of DD domain distortion. Additionally, fiber nonlinear effects are anticipated to be insignificant under the employed optical power levels. Therefore, the current ARoF implementation should result primarily in an additional link attenuation and power-budget penalty.

For the B2B setup, the antennas were directly connected to the SDR’s “TX” and “RX” ports, without involving the optical part.

The block diagram of the experimental setup is shown in Figure 4. The photos of the experimental setup, consisting of the CU and RAU with a static reflecting object, are shown in Figure 5.

Multiple measurements were made with a static object and a moving object—a rectangular metallic reflector, as depicted in Figure 6. The static object was located at a distance of $d=2.944 \mathrm{m}$ from the RAU 6 GHz horn antennas. The moving object was moved by the Kingroon KP3 (SHENZHEN KINGROON TECH CO LIMITED, Hong Kong, China) 3D printer motorized rail within a range of $1.78$–$1.8$ $\mathrm{m}$ from the RAU antennas. The setup for the moving object experiment is shown in Figure 7.

Performance evaluation was performed by transmitting $2 \times {10}^6$ bits to enable BER measurement at the order of $5\times {10}^{-7}$ (one error per measurement). This BER can be further reduced by applying FEC codes, such as the TPC, which can be seamlessly integrated into an OTFS system after QAM modulation in DD before IDZT, since the TPC structure is also two-dimensional [33]. It should be noted that TPC is only one possible choice for coded OTFS systems; alternative coding schemes, including LDPC and polar codes, can also be integrated within the same framework.

4. Experimental Results

4.1. Static Object

The received $-12$ $\mathrm{dBm}$ waveform of a single OTFS frame is shown in Figure 8. Averaged cross-correlation and cross-correlation distribution histogram from 14,000 OTFS frames between transmitted and received waveforms are shown in Figure 9.

Cross-correlation between the transmitted and received waveforms shows a pattern with multiple peaks at time lags when pilot pulses overlap. At these pilot overlapping time lags, the calculated correlation does not surpass half of the correlation value at zero time lag. Looking at the histogram, which is comprised of 14,000 OTFS frames, $90.55\%$ of the correlation values are in the range $[-10; 10]$.

4.1.1. Performance Depending on the Received Signal Power

Experiments with varying receiver frequency and transmitter power were conducted. The power of the received signal was calculated by averaging the whole received waveform using this formula:

$$P=10·{log}_{10}\left({\displaystyle \frac{1}{K}}\sum _{k=1}^K{\left|\mathit{x}\left[k\right]\right|}^2\right) ,$$

(8)

where P is the calculated signal power in $\mathrm{dBm}$, $\mathit{x}$ is the received signal vector, K is the number of samples in the signal $\mathit{x}$, and k is the index of the sample.

BER was measured in the optical setup, where BER, depending on the signal power at the receiver, $P_{\mathrm{RX}}$, is depicted in Figure 10.

A gradual decrease in BER was observed with increased received signal power. Starting from $-24.582$ $\mathrm{dBm}$, the recorded BER was below the estimator’s capability, indicating no errors in the received $2\times {10}^6$ bits with the implemented OTFS system in the optical setup. The maximum transmitted power was $P_{\mathrm{TX}}=-2.4$ $\mathrm{dBm}$ with the SDR’s gain set at 87 $\mathrm{d}\mathrm{B}$, with received maximum power of $P_{\mathrm{RX}}=-12.05$ $\mathrm{dBm}$, meaning that the whole optical setup’s attenuation for the maximum tested power was $P_{\mathrm{TX}}-P_{\mathrm{RX}}=9.65$ $\mathrm{d}\mathrm{B}$.

4.1.2. CFO for Mobile Object Simulation

An explicit CFO was added to the receiver to simulate an ideal reflection from the moving object having a constant speed. CFO was applied in both optical and B2B setups, and the simulation without SDR. The results from these measurements are seen in Figure 11.

In this experiment, the performance of the optical and B2B setups was not compared, as the power of the signal at the receiver was unequal: $-12.054$ $\mathrm{dBm}$ and $-7.7$ $\mathrm{dBm}$ for the optical and B2B setups, respectively. However, the overall dynamics of BER are still visible, as, in both setups and the simulation, only an integer Doppler shift $\mathsf{\Delta }f$ of 142 $\mathrm{Hz}$ provides adequate communication performance, with fractional Doppler having the worst BER, rendering data transmission impossible. The demodulation in fractional Doppler is limited by the implemented LMMSE channel estimator, which increases BER to approximately $>{10}^{-1}$ at CFOs $\Delta f \ne \mathbb{Z}·F_{\mathrm{r}}$ $\mathrm{Hz}$. Therefore, the comparison between the optical and B2B cases in Figure 11 is qualitative rather than a strict BER benchmark between the two links. Nevertheless, both setups show the same overall OTFS behavior, namely reliable operation for integer Doppler shifts and degradation under fractional Doppler, confirming that the ARoF section preserves the proof-of-feasibility functionality of the implemented OTFS chain, while introducing an additional power-budget penalty.

4.1.3. QAM Constellations

QAM constellations after OTFS demodulation were taken to assess and compare the time spread of QAM symbols in optical and B2B setups at different Doppler shifts. QAM constellations for both optical, B2B setups and simulation with different Doppler shifts—added CFOs $\mathsf{\Delta }f$ are shown in Figure 12, Figure 13 and Figure 14, respectively.

All QAM constellation points stay in their respective quadrants, which corresponds to BER ≤ 5 × 10−⁷ or 0 errors in the experiment with varied CFO. The only exception is in Figure 12d, where slight positive and negative phase offsets for the optical link are observed.

4.2. Moving Object

In this experiment, the movement of a metallic reflector was controlled using a Kingroon KP3 3D printer, moving the reflector back and forth over a range of $1.78$–$1.8$ $\mathrm{m}$, as depicted in Figure 6. An accelerometer was attached to the build plate of the printer in the direction of the movement to get velocity data for comparison with the implemented system’s estimated velocity.

Using part of the DD matrix where OTFS pilots are located, it is possible to estimate the speed and direction of the object from which the pilot signal was reflected. First, normalization was applied to the channel, and the pilot was located by finding the maximum value. Then, the largest value between positive and negative Doppler taps, nearest to the pilot, is taken to be used as a second weight $w_2$ in the speed estimation with a weighted average:

$$vs.={\displaystyle \frac{{\sum }_{k=1}^2\left(w_k·v_{\mathrm{tap}k}\right)}{{\sum }_{k=1}^2{w}_k}} ,$$

(9)

where v is the estimated velocity of the object in $\mathrm{m}/\mathrm{s}$; $w_k$ is a weight from the normalized channel with $w_1=1$—pilot weight and $w_2$—weight of the nearest Doppler tap with maximum weight; $v_{\mathrm{tap}k}$ is the speed of a Doppler tap in $\mathrm{m}/\mathrm{s}$.

Equation (9) was applied to all received channels. As Equation (9) uses weights from two close Doppler taps, the resulting velocity can be less than the single Doppler tap velocity resolution $v_{\mathrm{r}}$ of the system. This estimation provides finer precision over the tap resolution. The estimated velocity of a moving metal plate from the start of the movement is shown in Figure 15.

At the time of the experiment, the metallic reflector was moved forward and then back with a constant speed and a constant movement time of 1 $\mathrm{s}$, which is also observed in the sensor and estimated velocity graph in Figure 15. The velocity graph clearly shows the direction of movement, with positive speed for moving forward and negative speed for moving backward from the antennas. The maximum absolute velocity of $\left|v\right|\approx 0.4$ $\mathrm{m}/\mathrm{s}$ was recorded by the implemented system, which is incorrect, an absolute average velocity of an object recorded by the sensor is 0.15 m/s. This can be explained by the implemented system’s bias towards negative velocity (backward movement from the RAU), as no such increase or decrease from the expected velocity is observed for positive velocities. This bias can also be clearly observed in the error distribution shown in Figure 16.

The amount of deviation from the expected velocity for ≤$-0.1$ $\mathrm{m}/\mathrm{s}$ is at $16.05\%$ and ≥$0.1$ $\mathrm{m}/\mathrm{s}$ is at $2.31\%$, with the remaining $81.64\%$ being in the range $(-0.1; 0.1) \mathrm{m}/\mathrm{s}$. Calculated RMSE is equal to $0.0839$ $\mathrm{m}/\mathrm{s}$.

Received DD channel normalized matrices in the ZP region are shown in Figure 17 and Figure 18 for estimated velocity when object was static and moving towards the RAU antennas, respectively.

Both examples of received DD matrices show estimated velocity in the range of the accuracy—calculated RMSE when Formula (9) is applied. In both cases, the majority of the pilot occupies 0 distance and velocity taps. The weight value in the distance (delay tap) beneath the pilot at 0 $\mathrm{Hz}$ is present due to a sample shift in the received signal, which is due to the synchronization error. This can be improved by the use of an advanced synchronization method or system zeroing at a known distance.

5. Discussion

Existing implementations of SDR-based OTFS [20,21,22,23,24,25,26,27,28] provide insight into OTFS and its implementation across different SDR platforms, though none include an optical link, and only some present the implementation and results for OTFS with ISAC.

From an integrated sensing perspective, an experiment with a moving metal plate confirmed that the DD domain in the OTFS system can be directly exploited for estimating movement velocity and direction. This estimation can be achieved using only channel information from the received signal with the application of DZT to transform the OTFS waveform from the time domain to the DD domain. The experiment with a moving object has also shown that the implemented system can be employed for low-complexity monostatic sensing tasks.

This study has also shown the limitations of the implemented OTFS channel estimator. Satisfactory performance with zero recorded errors has been observed with a static object without Doppler and with integer Doppler shifts, with degradation in performance in the case where fractional Doppler was present. This is consistent with the implemented LMMSE estimator at the receiver, which is not suited for fractional Doppler. LMMSE estimator also suffers from poor scalability, providing the time complexity $\mathcal{O}\left({\left(NM\right)}^3\right)$, which limits OTFS grid expansion to higher tap values and overall system bandwidth. Ideally, a channel estimator with low computational complexity and the ability to work in the presence of fractional Doppler should be implemented, with the proposed algorithms presented in [34,35,36].

The parameters used in the SDR implementation of OTFS (see Table 1) were primarily chosen for velocity estimation, so that it was possible to get results using available equipment in the laboratory for moving object simulation. It is possible to increase the bandwidth of the OTFS waveform, decreasing velocity resolution, which in turn will provide the ability to capture higher speeds of a mobile object, with a maximum possible absolute velocity $\left|v\right|>17.7$ $\mathrm{m}/\mathrm{s}$, which is the limit of the implemented system. This limitation also impacted the distance resolution, making it $d_{\mathrm{r}}\approx 6$ $\mathrm{k}\mathrm{m}$, which is impossible to achieve experimentally without a delay simulation with hardware components. As a result of such low distance resolution $d_{\mathrm{r}}$, with distance orders greater than the tested distances of $1.78$–$1.8$ and 3 $\mathrm{m}$, the ranging of an object was not performed. To get a true joint range-velocity estimation, OTFS signal parameters of bandwidth F, frame time T, and carrier frequency $f_{\mathrm{c}}$ should be optimized for the experimental scenario, so that both meaningful distance and velocity values are captured.

The maximum limit of the bandwidth, on which both resolutions depend, of the USRP B210 SDR is 56 MHz [37], which translates to a velocity resolution of $795.45$ $\mathrm{m}/\mathrm{s}$ and distance resolution of $2.68$ $\mathrm{m}$ for any $M\times N$ size OTFS system at 6 GHz carrier frequency. The single tap velocity value can then be further decreased to $42.55$ $\mathrm{m}/\mathrm{s}$ if $M\times N=128\times 256$ or $M\times N=256\times 128$ OTFS is used, meaning that an increase in the size of the frame increases velocity resolution, keeping distance resolution constant, as it is dependent on the signal and SDR bandwidth. For this real-time OTFS system realization, a dedicated FPGA-based implementation of OTFS demodulation algorithms should be developed, as SDR communication with a PC through the software layer adds performance overhead, limiting the possible size of an implemented OTFS DD domain and working SDR bandwidth.

Overall, the real-time SDR-based system presented in this paper should be regarded as a proof of the feasibility of an OTFS ISAC chain in an ARoF setup, rather than a fully optimized communication and sensing platform.

6. Conclusions

This paper presented a real-time OTFS-based ISAC system implemented on a USRP B210 SDR, which was experimentally validated and evaluated in both B2B and ARoF configurations. In the optical scenario, the RAU was connected to the CU through an 800 $\mathrm{m}$ SMF span, thereby demonstrating the practical feasibility of combining OTFS, wireless transmission, and optical fronthaul within a single experimental framework. The obtained results confirm that OTFS is not only a theoretically attractive waveform for high-mobility communication and sensing, but also a viable candidate for real-time implementation in distributed radio systems.

In the optical setup with a static object—a metal plate located ≈3 m from the RAU—no bit errors were recorded in $2 \times {10}^6$ transmitted bits once the received signal power exceeded about $-25$ $\mathrm{dBm}$, corresponding to a measured $\mathrm{BER}\le 5 \times {10}^{-7}$. Velocity and direction estimation of a moving object was successful, providing information on the velocity in real time with a calculated RMSE = 0.0839 m/s, with a clear demonstration of the acceleration and deceleration of the object.

The measured attenuation of the complete optical setup, equal to 9.65 dB at the maximum tested transmit power of −2.4 dBm, provides a reference for the power budget in future ARoF-assisted ISAC implementations using similar optical setups. The overall system concept and hardware architecture have been successfully validated, whereas additional performance gains can be achieved through further optimization and integration of a low-complexity fractional Doppler algorithm.

Future work should focus on algorithmic, software, and hardware improvements. On the signal processing side of the software, more advanced channel estimation and equalization techniques should be investigated to improve robustness against fractional Doppler. Multipath propagation and more dynamic motion scenarios should also be investigated, as those are part of signal propagation in V2V and V2X scenarios. A key direction for hardware improvement includes transitioning from the current real-time PC- and Simulink-based prototype implementation to a dedicated FPGA-based real-time architecture. The transition of the OTFS processing block to an FPGA platform would enable higher throughput and frame rates, with reduced latency and lower computational requirements for the user.