Traditional interference waveform generation methods require prior acquisition of parameter characteristics of the target signal， making it difficult to achieve favorable results when confronted with the complex and diverse communication signal patterns nowadays. To address this issue， this paper proposes a communication interference generation technology based on lightweight deep convolutional generative adversarial networks （DCGAN）. The DCGAN network is utilized to perform unsupervised learning on the features of the target signal without prior information， thereby generating interference waveforms that are highly similar to the target signal. Meanwhile， lightweight processing is applied to the network to reduce the number of network parameters. In this paper， simulation experiments are conducted on communication signals of various modulation types. The experimental results show that the lightweight DCGAN network reduces the number of parameters by approximately 38% without compromising the quality of the generated interference waveforms. When the jamming-to-signal ratio （JSR） is above 0 dB， the bit error ratio （BER） of the communication system affected by the generated interference waveforms rapidly approaches 0.5， which is far superior to the interference effect of traditional interference waveforms.