extra bits

[Bruce Watson]

Given two recording bit rates (max of 24 Mb/sec, 17 Mb/sec), both of which resolve 1080 x 1920. What do the extra bits buy? That is, what is better about the visual quality of the picture when recording at the higher bit rate?

Camera is a Canon HF200, native AVCHD format capture. I'm talking about the difference between MXP and FXP modes here.

[Larry Horwitz]

The original video captured by the camera's sensor is generated at a rate vastly higher than 24 megabits/sec limit of AVCHD recording. When the user selects MXP, FXP, or other compression / encoding rates, the choice causes the camera's processor to "throw away" more or less of the original content, in the process trading quality for recording time.

The most noticeable impact is typically on fast motion events in the scene. Also noticeable is very fine picture detail and subtle color detail, along with greater noise. The visual outcome is more or less noticeable based on the original scene content, since some scenes require essentially "zero" mbits/sec to display whereas others may require 100's of mbits/sec to really capture in full detail. A camera pointed at a solid color wall / sky / scene would take essentially no bits to update, whereas a black hockey puck moving erratically across a white ice rink would need as many bits as it can get. (This is because the motion estimation method used in mpeg strongly assumes smooth vertical / horizontal motion, and struggles with diagonal motion, with or without acceleration).

It is instructive to take a 24 mbit/sec camcorder and record the same scene at differing rates. Playback on a good quality monitor will reveal the degradation (or lack thereof) and allow you to judge for yourself.

Another useful method is to take screen captures from the 24 and 16 mbit/sec clips of the same scene, and then compare the fine details captured. I have done quite a bit of this myself when evaluating new camcorder purchases, and also when comparing video editing software.

Larry

[Ron Evans]

I shot a heavy snowfall when I got my NX5U and compared to my XR500. On first viewing I could see little difference but on close examination the snowflakes on the NX5U video were sharp and distinct whereas on the XR500 there was just a little motion blur. This is a case where almost all the scene was changing all the time so the extra bits with the NX5U at 24mbps was clearly better than the 17 of the XR500.

Ron Evans

[Bruce Watson]

This is interesting.

My understanding is that AVCHD produces a variable bit rate depending on the motion of the scene. So if nothing is changing, the bit rate is very low. If there's lots of motion, the bit rate goes up. Yes? If that's the case, all the two shooting modes do is to put a cap on the bit rate. One at 24 Mb/sec, the other at 17 Mb/sec.

So... consider an interview. Two people, sitting. Not much motion. The bit rate will be fairly low regardless of the setting. Actually, it should be identical with either setting, yes? But if filming an olympic level ping-pong match, with lots of very fast motion, the higher setting provides a higher quality picture.

Am I correct in the above?

[Ron Evans]

Fundamentally yes but it isn't a simple cap. The algorithms to use available bandwidth will be different for the 17 and 24 mbps rates.

Ron Evans

[Tom Gull]

Originally Posted by Bruce Watson

This is interesting.

My understanding is that AVCHD produces a variable bit rate depending on the motion of the scene. So if nothing is changing, the bit rate is very low. If there's lots of motion, the bit rate goes up. Yes? If that's the case, all the two shooting modes do is to put a cap on the bit rate. One at 24 Mb/sec, the other at 17 Mb/sec.

So... consider an interview. Two people, sitting. Not much motion. The bit rate will be fairly low regardless of the setting. Actually, it should be identical with either setting, yes? But if filming an olympic level ping-pong match, with lots of very fast motion, the higher setting provides a higher quality picture.

Am I correct in the above?

I totally buy that there are times where the higher rate will produce visibly better video, but I'm thinking you might have to work pretty hard at it sometimes. There was a nice video posted using a CX550V showing the four different rates of recording against the same subject, typically in a four-way screen. It filmed all sorts of things that I would have thought looked different - running water, things blowing around, etc. I wasn't familiar with the Sony labels and that was what was onscreen to identify each clip. I leave my Sony cam on 17Mbs all the time and couldn't tell you which two letters represent that level. I struggled hard to figure out which quadrant represented which rate visually, but finally just gave up. The differences were very subtle and there were cases where I thought I had picked out the higher rate but didn't necessarily think it looked best. As with JPEG, we sometimes need a lot less than what could be there to see something as subjectively "good" or "beautiful".

I could easily see the difference between 1920x1080 and 1440x1080 video clips, and strongly preferred the former. I've had little luck telling the difference between the two highest bitrates to date.

[Larry Horwitz]

I have also had difficulty justifying the extra bitrate given the very similar performances of 17 and 24 mbit encoding.

Moreover, it sadly appears that few NLE software programs or AVCHD authoring programs can really handle 24 mbit/sec properly, and smart rendering (a feature I very strongly prefer) is not offered except on rare occasion for 24 mbit AVCHD. Thus my recommendation for the present time is to stick with 17 mbit shooting for most people and most circumstances.

Like most things in life, there are diminishing benefits which are disproportionately expensive or produce marginal improvement once a certain level is reached. Those of us who have pursued high end audio gear, very fine lenses, exceptional wines, etc. know that the first 90% of optimal comes a lot cheaper than the last 10% (the 90% Rule!), and h.264 with the cameras and sensors and lenses offered to consumers in the $1000 price range hit the sweet spot very well. A true 3 CCD/CMOS camera with larger sensors may justify the extra (24 versus 17) bitrate and offer a greater jump in image quality, but 17 does a really nice job for almost all typical purposes consumers need in these medium to low price camcorders most of us are using / discussing here.

Even the better DSLRs with much higher h.264 bitrates (30-50 mbits/sec) do not (to my eyes) look profoundly better in terms of fine detail or motion artifacts despite having twice the limit of AVCHD camcorders.

Larry