This contribution proposes a method for reducing bit rate fluctuations for achieving low end-to-end delay with random access capabilities. The method is referred to as distributed decoding refresh (DDR) initially proposed in contribution JVET-N0116 [1]. The key idea is to distribute IDR pictures over multiple inter-coded pictures, and such an IDR picture therefore cannot be used by these inter-coded pictures, except for the last one, for inter prediction reference. Such an IDR picture is referred to as a distributed IDR (DIDR) picture. A DIDR picture is not for output. Rather, the inter-codded associated DDR picture, which is a different representation of the same source picture, is for output. It is asserted that the DDR-based approach can achieve rate smoothness while higher coding efficiency than GDR-based approaches, because no motion restrictions as in GDR are needed. Compared to usual IRAP picture based coding, the DDR-based approach needs to encode one additional inter-coded picture for each random access period, and the last of the multiple inter-coded pictures over which the VCL NAL units of the DIDR picture are distributed is less efficient than usual due to that it cannot use the DIDR picture for reference, while the DIDR would often be a number of pictures away in POC distance. The proposed detailed spec text changes are provided in an attachment. Simulation results for CE11-4 was obtained according CE11 specific test conditions [2]. Results reportedly show that the average luma BD-rate overhead is 3.5% in LDB configuration. The influence on encoding/decoding time is insignificant.