The paradoxical evolution of sensor SNR over timeSunday November 16 2008 Sensor Insight
DSLR sensor technology has made significant advances over the past several years, with full-frame sensors outpacing the progress of APS-C sensors. DxO Labs measurements show that advances in pixel and RAW conversion technologies have compensated for the decrease in signal-to-noise ratio (SNR) brought about by the proliferation of smaller, less light-sensitive pixels in DSLRs. Overall, today’s high-resolution cameras produce higher-quality images than low-resolution cameras when viewed or printed under the same conditions.
When most photographers think of “noise,” they think of signal-to-noise ratio (SNR). In this Insight, we investigate how advances in sensor technology have impacted DSLR image quality with respect to SNR over the past few years.1 Specifically, we explore the impact of the megapixels race, and how technological improvements compensate for the inevitable loss of quality when using smaller pixels. We also examine the resulting overall gain when printing pictures in a fixed format.
The above graph shows the SNR as a function of the release date for 30 DSLRs (30 blue dots) and the average trend (black line).
X axis: Camera release date.
Y axis: SNR for gray 18% (mid-dynamic range) measured at real ISO200 and expressed in dB.2
We would expect the SNR to improve over the years, and indeed, a great deal of research has gone into introducing many improvements between each pixel generation. This said, there has been, surprisingly, an average loss of 1dB in SNR over the past five years! How can we explain this paradox? Should we conclude that image quality has degraded in the past 5 years?
The fact is that while many improvements have been made to pixel design, pixel sizes and quantities have also changed a great deal.
The above graph shows that the number of pixels has roughly doubled over the past 5 years. This increase has been achieved by reducing the pixel pitch by a factor of the square root of 2 (~1.4). Since the pixels are smaller and therefore less sensitive to light, they show mechanically a smaller SNR. Further, the 1dB loss observed in the first graph shows that new pixel technologies have roughly compensated for the mechanical quality loss inherent to pixel size reduction (along with the increased of number of pixels).
1 Findings in this Insight are based on working at real ISO=200. Camera rankings will differ when looking at ISO200 versus other ISOs (e.g., ISO1600), but as the objective of this paper is not to compare particular cameras, but rather to reveal tendencies, the chosen ISO is not of great importance.
2 To understand the SNR scale, remember that a 3dB difference between two cameras is roughly equivalent to a difference of one f-stop. That is, if camera A at ISO200 has an SNR of 3dB or larger than that of camera B, then camera A at ISO400 will have approximately the same SNR as camera B at ISO200, meaning that camera A will out-perform camera B by one stop.