In digital spread spectrum encoding, a stream of data bits is combined with deterministic noise to produce stream of encoded bits. These encoded bits have the desirable feature of appearing very noisy -- up to the repeat length of the noise source there is no apparent structure in the encoded data stream. This is desirable, because this stream of bits can then be transmitted in such a way as to completely utilize the bandwidth of the transmission channel.
For instance, assume you want to send eight bit values, and have 256 discrete frequencies in a channel. A simple transmission scheme would be to represent each byte as a pulse on its corresponding frequency. If you were sending ASCII text repressing this document, you would find that some frequencies were used much more than others (in fact, you would be using only one half of your channel since ASCII values are always between zero and 127). This would be a poor use of your spectral real-estate.
If on the other hand you were to spread your data using an LFSR, you would find that all of your frequencies were being used equally. Thus, you would be making maximal use of the channel, and the power spectrum across the discrete frequencies would be flat.