I ran some tests to measure this directly. I created a 5.1 test file with a 1 kHz sine at -20 dBFS on each channel in turn (60 Hz for the LFE), followed by a segment with L, C and Ls playing together. I placed it in a stereo sequence, exported a stereo WAV for each 5.1 Mixdown Type setting, and measured the results with Amplitude Statistics in Audition.
You're right that Premiere Pro doesn't use the standard downmix equation. It gives each channel a fixed weight:
- Left and Right: 1
- Center: 2/3, sent to both sides
- Left and Right Surround: 4/9
- LFE: 4/9, sent to both sides
It then divides by the total weight of the channels included in the selected 5.1 Mixdown Type. This means the included channels always add up to exactly unity, so the stereo output can't clip even if every channel peaks at the same time.
These are the resulting coefficients:
| 5.1 Mixdown Type | L / R | C | Ls / Rs | LFE |
|---|
| Front Only | 3/5 | 2/5 | 0 | 0 |
| Front + Rear | 9/19 | 6/19 | 4/19 | 0 |
| Front + LFE | 9/19 | 6/19 | 0 | 4/19 |
| Front + Rear + LFE | 9/23 | 6/23 | 4/23 | 4/23 |
For Front + Rear, that gives:
Lo = (9L + 6C + 4Ls) / 19
Ro = (9R + 6C + 4Rs) / 19
Compared with the standard equation, the center sits about 3.5 dB below the front channels instead of 3 dB, and the surrounds sit about 7 dB below instead of 3 dB. The normalization explains the overall drop you noticed. With Front + Rear, the front channels come out about 6.5 dB lower than the source.