I. Introduction
Advancements in Internet of Things (IoT) have heightened the demand for ultra-low-power communication technologies, with backscatter communication emerging as a promising solution owing to its minimal power requirements [1–3]. Backscatter communication operates by reflecting and modulating existing electromagnetic (EM) waves instead of generating new signals, making it ideal for battery-free or energy-constrained applications.
Traditional radio frequency identification (RFID) systems represent the most common application of backscatter technology, enabling identification and data transmission with minimal power consumption. However, since conventional backscatter systems require dedicated antennas for effective communication, they present design challenges for miniaturized IoT devices. Recent innovations in antenna-less backscatter techniques have addressed this limitation by employing existing circuit components and printed circuit board (PCB) structures as unintentional radiators, thus eliminating the need for separate antenna designs [4].
Notably, a significant challenge encountered in monostatic backscatter systems, where the transmitter and receiver are co-located, is self-interference (SI). The high-power transmitted signal leaks directly into the receiver, overwhelming the weak backscattered signals from the tag (modulated reflection). This interference severely degrades system performance by limiting both operational range and communication reliability, underscoring the broader vulnerabilities faced by electronic systems from EM phenomenon, which may even allow intentional data manipulation [5].
Previous solutions proposed for SI cancellation typically involved either brute force tuning (manual or automated) or predefined cancellation paths. However, these methods, which entail sweeping amplitude and phase parameters, are time-consuming and cannot be adapted to dynamic environments, thus exhibiting limited effectiveness in practical deployments. Since backscatter systems must measure weak reflected signals, their primary objective is to eliminate the strong returned transmitted signals that overwhelm the receiver. To achieve this, optimization is necessary because the SI signal phase may appear randomly upon each system initialization, thus requiring cancellation parameters tailored to each operating condition.
This paper introduces an efficient automated SI cancellation system that uses a bi-stage genetic algorithm (GA) to efficiently navigate the theoretically infinite amplitude–phase parameter space, demonstrating clear advantages over conventional brute force methods. Central to our optimization process is a fitness function for maximizing the high–low centroid ratio (HLCR). Derived from downconverted baseband I/Q components, HLCR refers to the ratio of the high-state to low-state signal centroid magnitudes in the I/Q constellation. A higher HLCR signifies improved separation between the desired signal states and residual interference, which is crucial for reliable demodulation.
II. Previous Experiments and System Configuration
Our previous research established the feasibility of antenna-less backscatter communication by identifying optimal mounting positions for nonlinear components (e.g., pHEMT ATF-54143) on PCB ground (GND) planes [4]. This was achieved based on the “scattering impedance” which can be defined as follows:
We then implemented the “hot-spot” estimation method by considering various GND configurations (such as isolated planes or those with narrow bridges) to maximize the backscattering ratio, achieving a 29.68 dB change in the return loss ratio at 1.745 GHz when switching the transistor state [4].
Building on this foundation, in the current work, SI removal techniques were evaluated using the controlled test environment depicted in Fig. 1. Notably, an Arduino Mega 2560 was selected as the device under test (DUT) in this study due to its open-source nature, which facilitated its integration with our test equipment.
For the monostatic configuration depicted in Fig. 1, SI occurred when the transmitted signal from Signal Generator #1 directly leaked into the oscilloscope through the circulator path (Port 1→Port 3), thus overwhelming the weak backscattered signal from the DUT. This SI signal can be mathematically expressed as follows:
where αiso represents the circulator’s isolation loss coefficient and A1 is the amplitude of the original transmitted signal.
Meanwhile, after passing through the circulator, the backscattered signal from the DUT can be formulated as follows:
where ΓDUT refers to the reflection coefficient of the DUT (which varies based on the transistor’s state), αcirc denotes the circulator’s insertion loss, αTEM signifies the TEM cell propagation loss, and φDUT is the phase shift introduced by the DUT.
To cancel out the SI, a cancellation signal from Signal Generator #2 is introduced, expressed as follows:
where A2 and φy are the amplitude and phase parameters, respectively, that need to be optimized to achieve the following:
Here, β denotes the coupling coefficient of the RF coupler and φSI is the phase of the SI signal.
In our initial experiments, we implemented brute force to sweep both the amplitude and phase parameters, as shown in Table 1, to identify the optimal configuration for interference cancellation. Fig. 2 presents a zoomed-in illustration of the results of this sweep, showing two clear optimal points where maximum HLCR of the received signal was achieved. Nonetheless, while this approach successfully identified the optimal parameters, the process took approximately 7,200 iterations to complete, making it impractical for real-world applications in which environmental conditions may change rapidly.
III. Bi-stage Genetic Algorithm Optimization
The choice of employing a GA in this study is motivated by its demonstrated ability to efficiently solve complex multi-dimensional optimization problems related to wireless communication systems, such as optimizing antenna position for enhanced network performance [6]. To overcome the limitations of brute force parameter sweeping and the challenges of applying conventional cancellation methods to antenna-less systems, we developed an automated optimization system using a bi-stage GA optimization approach. The proposed method is capable of efficiently searching the two-dimensional parameter space to identify optimal settings for SI suppression.
1. Bi-stage GA Structure
The proposed GA is structured as a bi-stage process—operating across two distinct optimization stages—as illustrated conceptually in Fig. 3. Stage 1 involves global optimization, which performs a broad search of both amplitude and phase parameters to quickly identify the optimal region. Once the optimal amplitude region is detected, Stage 2 fixes the amplitude to the best value identified and fine-tunes only the phase parameter, effectively reducing the search from a two-dimensional to a one-dimensional space for precise convergence.
2. Fitness Function based on HLCR
Guidance for the search was provided by a fitness function carefully designed to quantify the effectiveness of any given set of cancellation parameters (A2, φy). Departing from methods that directly minimize a residual SI envelope peak, we adopted a strategy that focused on maximizing the separability of the two primary signal states within the received I/Q constellation after cancellation. This objective was realized by calculating the HLCR.
Notably, a single fitness evaluation encompasses several steps. For a candidate pair of cancellation signal amplitude and phase values determined by the GA and applied via Signal Generator #2, the evaluation process begins with an oscilloscope capturing the resulting composite signal at the receiver input. This signal usually comprises the desired backscattered component, any residual SI, and ambient noise.
The captured time-domain waveform then undergoes I/Q demodulation, which transforms it into its complex baseband representation, xbb(t) = I(t) + jQ(t) [7]. Subsequently, low-pass filtering is implemented to remove out-of-band noise and higher-order mixing products, yielding a cleaner baseband signal.
In this study, we first computed the magnitude, |xbb(t)|, of each complex baseband sample to identify their distinct signal states. The probability density function (PDF) of these magnitude values was then estimated using the histogram method to reveal the distribution of the signal amplitudes. Fig. 4 shows the PDF of the magnitude before and after cancellation. We anticipated that a successfully cancelled bi-stage modulated signal would exhibit bimodal magnitude distribution. Therefore, the peak search function was employed to locate the two dominant peaks within the magnitude PDF, which were assumed to correspond to the average magnitudes of the low and high signal states.
Fig. 5 demonstrates the impact of optimized cancellation on the signal states within the I/Q constellation. The plot shows the distribution of the I/Q samples before and after cancellation. Before cancellation, the strong SI results in signal clusters or centroids that are significantly displaced from the origin, making the centroids of the two signal states either difficult to distinguish or far from an ideal state. After effective cancellation, the centroids are driven substantially closer to the origin of the I/Q plane. This transformation—achieving a low-state centroid with a small magnitude and a clearly separated high-state centroid—is precisely what the HLCR metric was designed to quantify.
Once the two characteristic magnitude levels were identified, a threshold was derived from them to segment the original complex baseband samples xbb(t) into two groups—those belonging to the low magnitude state formed one group and those belonging to the high magnitude state formed the other. The N complex centroid, representing the I/Q vector, was then calculated for each group, yielding the following equations:
Finally, the fitness score for the cancellation parameters A2, φy was computed as the HLCR, as follows:
The GA’s objective was to discover the A2, φy pair that maximized this fitness. Notably, a higher fitness value signifies a greater distinction between the two states, implying that the magnitude of the low-state centroid, |Clow|, is substantially smaller than that of the high-state centroid, |Chigh|. While this approach does not explicitly force |Clow| to reach absolute zero, maximizing the HLCR strongly encourages solutions where the low state is significantly attenuated compared to the high state. This enhanced separation is beneficial for the reliable demodulation of the signal by clearly distinguishing its two levels. Furthermore, by relying on centroids, which encapsulate both the average amplitude and phase (I/Q) characteristics of each signal state, this statistical approach offers inherent robustness against instantaneous noise and variations in data patterns within the backscattered signal. This robustness is crucial for practical deployments because it enhances the system’s resilience to minor environmental fluctuations and allows for extending the application of the proposed method to more complex modulation schemes and dynamic channel conditions where consistent signal separability is crucial.
The entire GA process, encompassing population management, genetic operators, and real-time interfacing between the signal generators and the oscilloscope, was fully automated by running a MATLAB script on a control PC. This automation achieved using standard interfaces, such as GPIB, TCP/IP, and USB, as required, facilitated the adaptability of the system.
IV. Results and Performance Analysis
In this section, the effectiveness and efficiency of the proposed bi-stage GA-based SI cancellation system are demonstrated through experimental results.
Fig. 6(a) presents a comparison of the received time-domain signal before and after GA-optimized SI cancellation. Initially, it is observed that strong SI completely masks the weak backscattered signal. After cancellation, a significant reduction in signal amplitude is observed, indicating substantial SI suppression.
The convergence behavior and efficiency of the bi-stage GA are crucial for its practical applicability. Fig. 6(b) compares the fitness evolution to the number of evaluations required when using the GA and a simulated brute force approach. The GA (blue line) demonstrates a rapid increase, achieving a high fitness level within a relatively small number of evaluations. In contrast, the systematic brute force sweep (orange line) exhibits a considerably slower stepwise improvement in fitness. Although it eventually finds the improved regions, it requires a significantly larger number of evaluations. This comparison highlights the GA’s superior efficiency in navigating the multi-dimensional parameter space to identify optimal SI cancellation settings compared to exhaustive search methods.
Further insights into the GA’s optimization process were attained by investigating the evolution of the I/Q centroids. Fig. 7 presents a three-dimensional perspective of the optimization process. The in-phase (I) and quadrature (Q) coordinates of all the low-state centroids from all generations are plotted on the X-Y plane, with the corresponding fitness represented on the Z-axis. In the Z-axis, a logarithmic scale—specifically, each fitness was divided by the maximum fitness—was utilized to effectively visualize the progression of fitness improvement. Each point, colored according to its generation, tracks the refinement in centroid position caused by the GA while simultaneously increasing the system’s fitness. The centroids with the maximum fitness values for each generation are connected by arrows, clearly showing the GA’s ascent on the fitness landscape, converging toward regions in the I/Q plane where the HLCR is maximized and achieving optimal SI cancellation.
V. Conclusion
This paper introduces an efficient automated method for SI cancellation in antenna-less backscatter systems by focusing on distinguishing the two signal states generated by impedance modulation. A bi-stage GA is proposed to optimize the amplitude and phase of the cancellation signal, guided by a novel fitness function—the HLCR—derived from the I/Q constellation. The HLCR serves to maximize the separability of high-state and low-state signal centroids, thereby enhancing demodulation reliability.
The study results showed that the bi-stage GA rapidly converged to enable optimal SI suppression, significantly outperforming the brute-force sweep method in terms of efficiency. Furthermore, visualizations confirmed effective parameter space navigation and SI mitigation by the GA, leading to a clear time-domain waveform. Overall, the proposed method offers a practical solution for robust antenna-less backscatter, especially in dynamic environments. Future work may focus on adapting this approach to more complex modulations and refining the fitness function to achieve enhanced performance.
PDF Links
PubReader
ePub Link
Full text via DOI
Download Citation
Print
