Non orthogonal multiple access (NOMA) is being considered a key technology for future communications. In this paper, we provide an analytical characterization of downlink NOMA orthogonal frequency-division multiplexing (OFDM) in presence of nonlinear effects associated to a high power amplifier (HPA) modeled as a memoryless polynomial. Taking advantage of the Bussgang approximation and intermodulation products (IMPs) tools, we present a generalized framework to characterize the nonlinearity for digital NOMA-OFDM systems with Nyquist- sampled signals and oversampled signals. An accurate estimation for the power spectral density (PSD) and signal-to-distortion ratio (SDR) for each user is developed and discussed. Moreover, we introduce a joint iterative approach for distortion compen- sation and interference cancellation suitable for scenarios with strong nonlinear effects. Our performance results indicate that the proposed technique allows significant bit error rate (BER) and complexity improvements, when compared to conventional schemes.