Main features Tubeless steel condensers STFT
Exclusive LU-VE technology – Oval refrigerant-fluid conduits
In the STFT tubeless condensers, it is the fins themselves which form the conduits of the fluid: the high, oval “funnel”-type fin collars are inserted one into the other and copper brazed in an inert atmosphere furnace. This method of construction gives the highest possible conductivity, as the gas virtually flows through the fins.
Very compact structure for a high ratio of power rating to volume.
These condensers can be supplied on request with plastic fan shrouds (model CF).
Spectrum Sensing Based on STFT-ImpResNet for Cognitive Radio
pectrum sensing is a crucial technology for cognitive radio. The existing spectrum sensing methods generally suffer from certain problems, such as insufficient signal feature representation, low sensing efficiency, high sensibility to noise uncertainty, and drastic degradation in deep networks. In view of these challenges, we propose a spectrum sensing method based on short-time Fourier transform and improved residual network (STFT-ImpResNet) in this work. Specifically, in STFT, the received signal is transformed into a two-dimensional time-frequency matrix which is normalized to a gray image as the input of the network. An improved residual network is designed to classify the signal samples, and a dropout layer is added to the residual block to mitigate over-fitting effectively. We conducted comprehensive evaluations on the proposed spectrum sensing method, which demonstrate that—compared with other current spectrum sensing algorithms—STFT-ImpResNet exhibits higher accuracy and lower computational complexity, as well as strong robustness to noise uncertainty, and it can meet the needs of real-time detection.
Keywords: spectrum sensing; residual network; short-time Fourier transform; cognitive radio
With the advent of the 5G era, the lack of spectrum resources has become a realistic problem that is inevitable . Spectrum sensing is of vital importance to the optimization of the utilization rate of spectrum resources, and has become the key technology in cognitive radio In cognitive radio, the secondary user (SU) is allowed to access the spectrum dynamically and randomly without interfering with the primary user (PU) . The main task of spectrum sensing is to explore spectrum holes in order to increase the usage of spectrum resources.
The traditional spectrum sensing methods can be broadly categorized into energy detection (ED) , matched filter detection , cyclostationary feature detection , waveform-based sensing , and covariance-based detection , etc. However, the pre-defined threshold set by the traditional method has a dramatic influence on the detection probability. With the continuous development of machine learning techniques, the method of realizing spectrum sensing is migrating gradually from conventional statistical methods to machine learning ones. Nowadays, deep learning methods are becoming more and more popular to train spectrum sensing models to classify signals, which improves the detection probability of spectrum sensing, and the model is optimized to approach the pragmatic application level.
At present, some commonly used machine learning methods such as support vector machines (SVM), artificial neural networks (ANN), long-term and short-term memory networks (LSTM), and convolutional neural networks (CNN) have achieved partial success in spectrum sensing. Chen et al. proposed a SVM-based spectrum sensing algorithm to recognize the PU signal by training SVM classifiers based on the energy vectors sampled from SU. Supervised learning and unsupervised learning algorithms such as the naive Bayes classifier, SVM, and hidden Markov model are compared in terms of classification accuracy in , in which the experimental results show that the performance of the SVM algorithm exceeds previous ones.
However, as the SVM algorithm uses the time-consuming quadratic programming to solve support vectors, it exhibits high computational complexity in the training process, along with relatively low detection efficiency. Some researchers have proposed new spectrum sensing methods based on ANN and its variants, and also combinations with traditional methods.
Tang et al. used energy detection and cyclostationary characteristics to train an ANN model for spectrum sensing, which combines the advantages of energy detection and cyclostationary feature detection while keeping a low computational complexity. The normal likelihood estimation scheme is employed in to input the signal energy detected by the energy detection method to the ANN spectrum sensing model, and the experimental results were better than those provided by the straightforward energy detection methods. In [16], the decision level fusion is introduced to ANN during the spectrum sensing of cooperative users. The decision of each SU achieves the global decision in the fusion center, which improves the detection probability and reduces the false alarm probability in the meantime. However, although the sample data amount increases, the training process is still prone to over-fitting due to the simple design of the network structure, which limits the accuracy of ANN algorithms. To this end, many researchers empower the communication signal recognition tasks by deep learning techniques, regarding signal recognition as a classification problem. Dong et al.
extracted both cyclostationary and energy features from noise signals and PU signals, respectively. These features are input to the CNN spectrum sensing model, and the detection can be made by judging whether the frequency band was occupied. However, these features are insufficient to accurately describe the real environment. Subsequently, Pan et al.
proposed a cognitive radio spectrum sensing method for orthogonal frequency division multiplexing (OFDM) signals based on the integration of deep learning and the cyclic spectrum. This method analyzed the cyclic auto-correlation characteristics of OFDM signals and the cyclic spectrum obtained by the time domain smoothing fast Fourier transform accumulation algorithm as the input of the CNN model. However, this algorithm is not generalized, as the model is merely tailored to OFDM signals. Wu et al.
established a signal modulation recognition model based on CNN-LSTM, which can identify as many as 12 types of signal modulation modes simultaneously. CNN was used to extract the characteristics of the signal space automatically, and then the LSTM network was exploited to extract the time correlation of the extracted signal. The authors in proposed a deep belief network architecture, which achieves better data transmission through the selected path. A spectrum detection network based on deep learning is proposed in to identify the channels; it achieved good results in a low signal-to-noise ratio scenario. Nevertheless, this method is only suitable for specific scenarios, and is not generalized as well. Chen et al.
proposed an STFT-CNN spectrum sensing method that utilizes short-time Fourier transform (STFT) to preprocess the signal to make full use of the time-frequency domain information of the signals, and designed a CNN network to classify the signal. This method is a milestone in spectrum sensing. However, the CNN network only contains a single convolution layer, which limits the ability of feature learning. The network’s performance can be improved by increasing the network depth, especially for low signal-to-noise ratio (SNR) spectrum signals. However, introducing excessive network layers leads to network degradation. Specifically, the classification accuracy increases with the deepening of the network layers at the beginning.
As the network continues to deepen, the accuracy drops sharply after reaching the saturation point. The reason for the network degradation is that with the deepening of the network, the gradient correlation between the shallow network and deep network becomes weak, and a loss of information occurs. To capitulate, traditional spectrum sensing methods have a low utilization rate of signal features with limited feature information extracted, and have a lot of space for the improvement of the accuracy of the spectrum sensing.
view of the aforementioned issues, we propose a novel spectrum sensing method, named STFT-ImpResNet, based on STFT and an improved residual network (ResNet). Specifically, the signal samples are preprocessed by STFT, and an improved ResNet (ImpResNet) is designed to classify the signal samples.
The main contributions of our paper are as follows:
We combined Short-time Fourier transform and a residual network innovatively, and proposed STFT-ImpResNet for spectrum sensing. To the best knowledge of the authors, this is the first time that a combination of STFT and ResNet has been introduced to spectrum sensing.
We customized a deep learning network structure to achieve a good trade-off between accuracy and computational cost in the context of spectrum sensing. We especially simplified ResNet by replacing the fully connected layer with a global average pooling layer to integrate global information. In addition, a dropout layer was added into the improved residual block to prevent over-fitting effectively.
We conducted comprehensive experiments which demonstrate that the proposed STFT-ImpResNet algorithm outperforms the existing spectrum sensing algorithms on low signal-to-radio datasets.
The remainder of this paper is organized as follows: we explain the system model of spectrum sensing. The proposed STFT-ImpResNet Spectrum Sensing Algorithm is elaborated in followed by the extensive experiments in Section 4 which validate the superiority of STFT-ImpResNet in the balance of accuracy and detection efficiency.
0 Comments