Rqolqhdulw\\(vwlpdwlrqqlqq3rzhuu$psolilhuv%dvhggrq 8qghuvdpsohgg7hpsrudoo'dwd $evwudfw

Precise and representative power amplifier models are a key factor in order to simulate an entire communication system and essential in order to introduce linearization techniques (e.g., predistortion). This work analyses the distortion caused by power amplifiers driven by wideband signals. The traditional measures (AM-AM and AM-PM curves), characterizations and models (Saleh [1], Poza [2] and Abuelma’atti [3]), which are based on single tone amplitude sweeps are not capable of truly representing the device behavior due to the use of modern wideband digital modulations (OFDM, CDMA, QAM,...) with nonconstant envelope. Nowadays, analog to digital (A/D) converters have seen significant advances both in performance and speed, while prices have declined. Advances of digital signal processing are evident and the tendency to substitute analog signal processing by their equivalent digital counterpart is widespread. However, the well-known Nyquist limit represents a very serious obstacle preventing the application of DSP techniques at frequencies higher than a few hundreds of MHz. However, sampling at rates lower than the Nyquist limit, twice the maximum frequency, can still allow for an exact reconstruction of the information content of the analog signal if the signal is bandpass. This is the undersampling (or bandpass sampling) technique [4-6], that allows exact reconstruction of the original bandpass signal if the sampling rate is at least two times the bandwidth of the signal. An important practical limitation is that the A/D converter (or the Track&Hold) must still be able to effectively operate on the highest frequency component in the signal. Undersampling is a common technique in several applications, for example, radio receivers using digitization at the intermediate frequency or, in some cases, even at the radio frequency. In this paper we will use this technique for device characterization and develop polynomial models with memory from de available data. Several authors have presented polynomial memoryless models [7], tapped-delay-line models [8], Hammerstein models [9], Volterra models or Neural Network models [10] using input-output temporal measures in order to identificate nonlinear systems. The paper is organized as follows. In Section 2 the data Acquisition System using undersampling is described. In Section 3 several measurements of power amplifiers are presented. Section 4 deals with the polynomial models, and Section 5 presents the main results. Finally, in Section 6, the Conclusions are exposed.