The envelope X coordinates are values between 0 and 31. They are in units of 4 samples in the subband time domain buffers, so they represent sample numbers between 0 and 124. As the points are stored in ascending X order, some encoding method exploit that the X coordinates are monotonically increasing.
The first value is a plain 5-bit number. For all later points, the encoding depends on the magnitude of the previous X coordinate value. If it was below 15, the current point is directly encoded as 5-bit number. Otherwise, the value is delta-encoded relative to the successor of the previous X coordinate value, with as few bits as possible for the delta value. That means for values between 15 and 22, there are 4 delta bits to be able to reach 31 (with 3 data bits, the maximum encodable number is 7, so from 22 the highest value reachable with 3 bits would be 22+1+7 = 30). For 23..26, 3 bits are used and so on up to 30, where the delta (only 0 is valid) is encoded using 0 bits.
The first X value of each envelope is directly encoded. The following X values are delta-encoded using one of two huffman trees depending on whether there is an increasing or decreasing slope.
The X value for the first point is stored as difference to the X value of the first point in the master channel, assuming 0 if the master channel has no point; difference application wraps around after 31. This difference value is variable-length encoded. Each further X value that has an associated X value (with the same point number) in the master channel is encoded depending on whether it is an upwards or downwards step. For upwards steps, a bit indicates whether the X value should be cloned or it is stored like in mode 0, while for downwards step, the difference is stored to the master channel X value. The remaining template-less X values are encoded like in mode 1 on the master channel.
The envelope for band 0 is stored as in mode 0 (that means: if band 0 has no envelope, no envelope is stored that way!). The X coordinates of the later envelopes are stored as difference to the X coordinate of the corresponding point (with the same index) of the directly preceding envelope. Application of deltas wraps around. Different trees are used for increasing or decreasing slope of the envelope, the first point is defined to be handled like on a decreasing slope. In case that the directly preceeding envelope has no point with the same index, it is assumed as zero for the first point in the envelope or as the previously read X value of the current envelope for later points.
For each envelope in the slave channel having not more points than the corresponding envelope in the master channel, a bit is stored indicating loading like in master/mode1 or cloning like in slave/mode3. If the slave envelope point count exceeds the master envelope point count, the envelope is loaded as in master/mode1 without a mode select bit.
One base value (stored plain binary) is used for all coordinates in all bands. Every X coordinate is obtained by adding a second number to the sum of the base value and the current point number (0-based). This second number is also stored in plain binary, but with fewer bits (but the same bit count for all X coordinates in one channel). No wraparound happens, overflow makes the data invalid.
The X values for the envelopes in the slave channel are copied from the X values of the master channel. If an envelope in the slave channel has more points than the corresponding envelope in the master channel, the missing values are encoded near-directly as in master/mode0