A major problem affecting the design of data compression systems is that of employing a buffer of limited size and at the same time prevent uncontrollable loss of data due to overflow. One method of alleviating this problem is to employ an adaptive compression algorithm. With this design approach when overflow is imminent the compression algorithm is degraded which effectively reduces the input rate to the buffer. A method is proposed here, where by using a recirculating register as the buffer the recirculating data controls the input rate and hence the performance of the system. The system has been analysed for a Poisson input process, and simulated using synthetic patterns similar to that encountered on sonar displays. The results indicate that this form of storage is quantitatively similar to random-access storage but qualitatively superior due to the random nature of the losses. An experimental system has been built using dynamic MOS shift registers for the store and a simple run-length coding procedure.