In response to the escalating demand for high-performance, miniaturized, and
integrated radio frequency (RF) systems, this research explores the application
of the Zynq UltraScale+ RFSoC XCZU47DR chip in the realm of integrated RF
transceiver technology. An 8-channel, 4.8Gsps multi-channel distributed
collaborative spectrum sensing architecture has been designed, incorporating
lightweight IQ neural network, which comprises a convolutional layer, three
Bottleneck Units (BNU), a Global Average Pooling (GAP) layer, and a Fully
Connected (FC) layer. Notably, each BNU encapsulates one or two inverted
bottleneck residual blocks that integrate the concepts of inverted residual
blocks and linear bottlenecks. The parameter counts and computational complexity
associated with the convolution operation are significantly reduced to merely
11.89% of those required by traditional networks. The performance metrics of the
hardware circuit were validated through a constructed test system. Within a 2GHz
instantaneous bandwidth, the amplitude consistency between Analog-to-Digital
Converter (ADC) channels is less than 1dB, and the effective number of bits
exceeds 7.3 bits. Simulation results demonstrate that, at a Signal-to-Noise
Ratio (SNR) of -10dB with a false alarm probability of 0.5%, the detection
probability of the collaborative spectrum sensing algorithm reaches 91.13%,
marking a 6dB enhancement over conventional energy detection methods. This
achievement underscores the technology’s substantial advantage in boosting
spectrum sensing capabilities, providing novel perspectives for the design of RF
systems and the evolution of wireless communication technologies.