At a minimum, a spread spectrum communications system sends information over several frequencies or other separable means of communication. While the information being sent is redundant, the power levels of the frequencies, etc., are reduced in comparison with a single-channel message. The receiver extracts the content from signals present in the multiple channels.
"Frequency" is used as an example of a means of creating multiple channels, but frequency division certainly is not the only method. At a minimum, alternatives include time slots, or possibly synchronization code sequences.
The first patent on the technology was issued in 1942, jointly to the actress Hedy Lamarr and the composer George Antheil. Lamarr had the insight into multichannel transmission, and Antheil developed an electromechanical implementation based on a punched paper tape "player piano" control. Serious implementation, however, had to wait until electronics were sufficiently capable, in the 1960s. 
In the simplest case, when the individual channels are not secret, spread spectrum gives greater noise immunity to the transmission. If ten channels were used, and a noise burst only affected one or two of the channels, that common information can be extracted from eight channels gives a high assurance that the information is correct.
Spread spectrum, in combination with other techniques, also can provide considerable security. To use some arbitrary numbers, assume that twelve channels are available to send the information, and that a successful reception requires that the signal on four channels is consistent. The actual signal might be sent on six of the channels, with the other channels sending not random noise, but data that has some statistical commonality with the real data. In the present of apparently statistically meaningful material on multiple channels, selecting the true channels is difficult.
When spread spectrum is used for security, the signal is usually encrypted, making it much more difficult to find the true channels based on statistical properties; the false channels might receive ciphertext of random noise being sent through the same encryption as the true information.
In combination with frequency agility, spread spectrum becomes more powerful. In the previous example, even if the true and false channels were fixed, it would still take some effort to find the channels that correlate well. When frequency agility is used, however, the true and false channels frequently change, according to a synchronized pseudorandom scheme agreed between the sender and receiver. The number of channels used for true and false data might also change; if four channels are needed for receipt, as long as the signal goes out on at least four chanels, the number, as well as the slot, of true channels can vary.
- From Strapless to Wireless, Hedy Lamarr Foundation