Tests and settings on a Canon EOS C300
Tests were made on a pre-production sample of the Canon EOS C300. Clearly, the specification and performance of the camera may differ when it is released, and therefore any details and measurements in this document are subject to modification, when a production model is tested. Canon have made it clear that this is only their first offering of a ‘digital-cinema’ camera.
The camera has one large CMOS sensor (super 35 size, 24.6×13.8mm) and uses interchangeable lenses. It will be available with one of two lens mounts, the standard PL, and Canon’s EF range. EF lenses can be controlled from within the camera, while the PL mount has no electrical connections to the lens.
It records MPEG2-compressed video, and has HDMI and HDSDI output, but all the tests were made using the recorded MPEG2 signals, and analysed in software. It has a built-in monocular viewfinder, and connection ports for external LCD panel, control handle and other accessories.
The sensor is a single CMOS, total 4206×2340 photo-sites, of which a central patch of 3840×2160 is used for the video signal (the implications of this will be discussed in detail in the measurements section).. Recording is onto Compact Flash card (two slots) in MPEG-2, long-GoP, with MXF file format. Three bit rate options are available: 50Mb/s CBR (constant bit rate) at 4:2:2 colour sampling (1920×1080 or 1280×720), 35Mb/s 4:2:0 VBR (1920×1080 or 1280×720) and 25Mb/s 4:2:0 CBR (1440×1080 only). Thus it complies with broadcast requirements for bit rate and offers more economic rates for greater economy (the 25Mb/s option matches HDV format). At these rates, a 64GB card can record 160, 225 and 310 minutes respectively. In 1080 mode, both interlaced and progressive modes are available. Off-speed recording at fixed speeds from 12 to 60fps is possible. Recorded content is to 8-bit depth1; this is a limitation of the internal processing. HDSDI and HDMI outputs are also both 8-bit depth, although the data-stream is 10-bit. This does not appear to have any detrimental effect on the camera performance.
The EBU-approved 50Mb/s MPEG2 format is 4:2:2 coded at 10-bit depth. Although the camera records 8-bit content, it should be handled as 10-bit in post-production since this gives ‘foot-room’ for image manipulation without introducing colour contouring. No contouring was experienced during the camera tests.
Since the camera’s digital processor handles only 8- bit signals at the output, it may not be handling more than 10-bit data at the ADCs, and so non-linear ADCs make a lot of sense. It is clear that a better digital processor would result in significantly lower noise values in this camera. However, this non-uniform distribution of noise gives the pictures a more film-like appearance, where the pictures appear to be ‘more quiet’ than the numerical analysis alone would indicate.
The measured noise levels are not low enough to be limited by the 8-bit nature of the recorded signal, which has a noise floor of about -54dB. The camera’s performance appears to be well-matched to 8-bit output.
This camera performs well. Resolution is very well maintained and is refreshingly alias-free, far more so than other CMOS single-sensor cameras. Detail controls work well, and the factory settings are generally good. Noise levels are similar to those of 2⁄3” cameras with 3 sensors, as is sensitivity. Dynamic range is unusually high, at least 12 stops. Operating the camera at significantly higher gain produces more noise, but not dramatically so. The specified noise level of -54dB is achievable only at -6dB gain and with noise reduction switched on. Noise distribution is non-uniform, which gives the pictures a more film-like appearance.
Because of the large-format size of the sensor, iris diffraction should not start to be visible until the lens is stopped down to between F22 and F32. This, together with the integral neutral density filters, means that there should be about 14 stops of useful exposure control.
Performance at 720p is acceptable although not quite ideal, and can be improved a little by judicious use of the noise reducer. It is probably not essential to shoot at 1080 and use an external software or hardware down-converter, although the resulting spatial aliasing would probably be better by doing so.
For the full test click here: