Global Sources
EE Times-Asia
Stay in touch with EE Times Asia
EE Times-Asia > Processors/DSPs

In notebook webcams, less pixels means more performance

Posted: 04 Jun 2007 ?? ?Print Version ?Bookmark and Share

Keywords:integrated notebook webcams? broadband? megapixel webcams? megapixel imaging?

By Shone Tran
Cypress Semiconductor

With the recent proliferation of integrated notebook webcams, designers today face the challenge of balancing actual consumers need vs. marketing seesaw. Webcams risk heading down a path similar to digital still cameras (DSCs)the drive for more pixels (higher resolution) when there really isn't a compelling need for one. In reality, going down this route can severely degrade end-user experience because packing more pixels into smaller sensors translates into smaller pixels, which dramatically decreases low-light performance.

Currently, chat programs like AIM, Yahoo, Windows Live Messenger and ICQ do not support high-resolution web conferencing or video chatting because of very limited broadband infrastructure. Hence, webcam designers should not rush to design in the highest-resolution webcams where there is no real application need.

This article will suggest that notebook PC OEMs take a step back and ask themselves whether they are designing with the right considerations to optimize end-user experience. This paper will discuss key issues related to integrating webcams into consumer notebooks, including how to maximize the performance attributes that really affect image qualitysize, low light performance, resolution and frame rate, while keeping in mind the limited broadband infrastructure and the typical applications/environment in which consumers operate their webcams.

Rise of megapixel imaging
The average consumer was first introduced to the idea of megapixels when DSCs took off in the late 1990s toward the early millennium. At the time, the subject of conversation was on what DSCs could do and how convenient they were compared to traditional 35mm film cameras. Fast forward to 2004, the casual user had already become comfortable with using DSCs and was starting to compare one camera's megapixels to the next when shopping for a new camera.

At the beginning of this trend, migrating quickly from VGA to 1.3Mpixel, 2Mpixel and then to 3Mpixel cameras did benefit the end users because they would realistically print pictures in 4-by-6-inch or 5-by-7-inch film papers. As camera vendors started packing in more pixels than realistically needed (4Mpixel and more), consumers began to mistakenly associate the number of pixels with the actual picture quality of the camera.

Perhaps, this happened because it was a quantitative metric that consumers can grasp and compare, or it was simply the result of great marketing by DSC vendors, functioning as a tool for differentiating their products. Few consumers understand that owning cameras with more megapixels are only necessary if the pictures taken need to be blown up for editing or developing large prints. From the surface, this may not appear to be a big issue as consumers pay less for more megapixels over time, except now that they are brainwashed to take the same approach when evaluating integrated webcams in their notebooks. Notebook PC vendors are glad to feed the fire and give consumers what they were trained to look for. Ultimately, this can be detrimental to the consumer experience because of the degradation of image quality in higher megapixel webcams. Integrated notebook webcam designers must consider other important factors beyond megapixels to improve end user experience.

Dialing up to notebook cams
There are two typical characteristics about the environment in which consumers use integrated notebook webcams. First is over the Internet on instant messaging (IM) chat clients and second, in unfavorable lighting setups like the home or office. Designers must consider these characteristics when designing integrated notebook webcams.

Let's tackle first the Internet and IM environment. Notebook PC vendors are rushing to integrate webcams as online IM gains rapid popularity. AOL Instant Messenger (AIM), Yahoo! Instant Messenger (YIM), Windows Live Messenger and ICQ are the most popular IM clients that support video-based chatting over the Internet. While the Internet is powerful enough to connect users worldwide, it presents a major restriction to webcam performance: limited bandwidth. A corollary aspect to this is the fact that different users have varying connection speeds. IM clients take this into account by compressing already low-resolution video to manageable sizes to transport it through the Internet using the average connection speed.

When IM video conferencing over webcams was first introduced in 2003, the average consumer broadband connection speed was around of 384Kbps down/128Kbps up. To compare this in context with webcams, consider Apple's popular iChat software which supports AIM, ICQ, Jabber and .MAC protocols and uses H.264 compression to deliver MPEG-2 quality video over the web at half the data rate. Even with using such high compression technology, Apple's iChat requires a minimum of 100Kbps up/down bandwidth to squeeze VGA resolution video through the Internet at 30fps. At a glance, the 384Kbps down/128Kbps up average broadband speed appears to suffice. But factoring in the simultaneous usage of e-mail, web browsing and the physical distance between users across the globe, suddenly makes 384Kbps down/128Kbps up speeds barely sufficient. Over the past few years, average broadband speed has inched up to averaging between 384-768Kbps down and 128-384Kbps up, which is less than two times the bandwidth increase.

Meanwhile, integrated notebook webcams were introduced with VGA resolution by notebook PC vendors like Sony and Asus in 2004. Then come 2005, other notebook PC vendors began introducing webcams but skipping VGA altogether and immediately entering the market with 1.3Mpixel cameras. The additional bandwidth requirement increased around four times going from VGA to 1.3Mpixel. Comparing this to the increase of broadband speed adoption, which is about twice, integrated notebook webcams seem to be outpacing the broadband infrastructure in terms of bandwidth requirement. The advancement of integrated notebook webcams up the megapixel curve is simply not feasible.

While VGA and 1.3Mpixel are still the dominant resolution in 2006, notebook PC vendors are already designing in 2Mpixel cameras. Will consumers adopt the 2x or 3x bandwidth required within the next year or two? It seems unlikely, as broadband adoption overall is leveling off at 53 percent in 2006 and Internet service providers charge a tremendous premium for high upload speeds. Analysts predict that even if adoption rate increases, the speeds will stagnate or even decrease on average as competition drives the low-cost packages down.

Now, let's evaluate the typical environment where a user operates his webcam: in the home or at the office. The typical camera flash emits 2,000W of light to properly expose a picture. A typical home or office environment uses lighting around 100-150W light bulbs, perhaps with multiple bulbs. This order of magnitude difference shows that users are operating webcams in unfavorable lighting conditions.

Image sensor vendors do all sorts of tricks like decreasing frame rate to maximize integration time and using pixel binning to accumulate additional light from neighboring pixels to improve low light performance. This generally works when taking still image captures, but when capturing video, these methods severely hurt frame rate and affects picture clarity.

Choosing the right architecture
Satisfying mechanical constraints for webcam design is probably where the engineer really starts to choose a webcam architecture that meets the size constraints put on by the laptop bezel, while satisfying performance requirements. For example, given a very small bezel, perhaps low light performance will be compromised and the designer will need to choose an architecture that maximizes pixel area and pixel size.

Choosing the proper architecture can make or break a webcam design, as each has its pros and cons. Consider the following architectures:

Figure 1: Choosing the proper architecture can make or break a webcam design, as each has its pros and cons.

SoC + USB - This is a two-chip solution that uses an image sensor integrated with image processing (SoC) and a Hi-Speed USB peripheral controller. A particular benefit of this architecture is that the image sensor vendor knows their sensor and can optimize the ISP to best complement the sensor characteristics. A drawback of going with an SoC is the fact that sensors tend to be designed thin and use less metal layers to build the ISP logic. Thus, it may take more silicon area to realize the same ISP logic in an SoC, reducing pixel area and further limiting the mechanical constraints of a laptop bezel.

Image sensor + back-end chip (ISP + JPEG + USB) - This is also a two-chip solution, but the difference is that the image sensor stands alone while the image signal processing (ISP) portion is integrated (sometimes with JPEG compression) with the USB controller. This architecture allows you to choose a bigger sensor with larger pixels, while the designer can choose a different ISP for a given sensor and gain more flexibility. At the same time, this can be cumbersome for designers not verse in evaluating imaging pipelines. Likewise, a possible disadvantage is that a given pipeline may not be optimized for the different image sensors a designer might opt to use.

Image sensor + ISP + USB - This is a one-chip integrated solution which has currently gained little traction because it is limited to full-speed USB, which then ultimately limits the resolution. This solution is potentially limited to full-speed USB because high-speed USB uses extremely fast signaling, which generates significant heat. Since the USB is essentially built onto the same die as the pixel array, heat is a big issue in this solution. For those trying to differentiate their devices, using this architecture limits you to only one or two vendors.

Rendering crisp images
Resolution is another central issue in webcam design. As discussed, there is an obvious tradeoff between megapixels and frame rate and low light performance. A major consideration is whether there really is a need for video conferencing in high resolution.

Traditionally, higher resolution would enable blowing up an image to do editing or to print larger stills. These two factors are effectively moot points if the webcam is used for video capture. Consider the average laptop or LCD display, which run on 1,024x768. Such a display would not show an entire 1.3Mpixel picture. Consider also the usage model of web conferencing, where users are usually multitasking. This means users would not want a full screen video anyway, because they must access their e-mail, or view a document that they're discussing. Avoiding the megapixel marketing trap is the only way to deliver the best image quality and smoothest video over the current Internet infrastructure. Designers should ask themselves if offering more pixels will really improve the end user experience, or actually drag down the factors that really matter.

Figure 2: A 0.25-inch sensor has considerably less surface area than a 0.33-inch optical format sensor, so there's less light to work with.

The optical format attribute simply specifies the diagonal length of an image sensor. While a given resolution image sensor can have a range of pixel sizes, given the same optical format size (typically 0.33- or 0.25-inches), more megapixels inevitably translates into smaller pixels. Optical format is an important consideration, as notebook PC vendors are trying to fit webcams into the small bezel of a notebook LCD. Ideally, a designer will choose the largest optical format that will fit within physical limits, while using the largest pixel size at a reasonable resolution.

In terms of frame rate, the human eye generally sees fluid motion at 24fps. Typical VGA CMOS sensors found in webcams can capture video at 30fps in full light, but increasing integration time, for example, can easily reduce the frame rate below 20fps, where the human eye will perceive choppiness. Minimizing the need for integration time to improve low light will keep the camera from having to run lower frame rates to compensate for poor low light performance.

Meanwhile, low light performance a characteristic of webcams that is difficult to measure. Designing for maximum low light performance requires pinpointing the factors that most directly affect it. Pixel size is directly related to low light performance because in general, the bigger the pixel, the more light it can collect, which then improves exposure. Given a constant optical format, typical pixel sizes for VGA sensors are in the 5.6?m range, while 1.3Mpixel sensors are in the 2.8?m range and 2Mpixel sensors are in the 2?m range.

Optimizing design functions
USB throughput, heat and dynamic range are also attributes that are critical to webcam design.

Webcams are ubiquitously connected to PCs via USB connections and is the method of choice even in integrated notebook webcams. To ensure smooth video, the USB 2.0 specification allows for dedicated bandwidth via isochronous transfers. The maximum theoretical throughput using isochronous transfers is 24MBps. While Hi-Speed USB 2.0 is a standard, not all USB controllers are created equal and not all of them will produce robust, uninterrupted 24MBps isochronous transfers.

Choosing a programmable and reliable, high-speed USB peripheral controller ensures maximum frame rate and easy upgrading to accommodate the ever-improving offering of image sensors. Most importantly, USB becomes a bottleneck above VGA resolution. Theoretically, USB delivers uncompressed video at 30fps at VGA and 15fps at 1.3Mpixel resolutions. Squeezing more megapixels and higher data rate through USB requires adding hardware compression, which will inevitably drive up the BOM cost of the integrated webcam. Integrating an additional layer of compression, which will be uncompressed and then recompressed by the IM client will further degrade image quality.

Heat is an often subtle and overlooked consideration is the problem of heat and its effect on image quality in a webcam. Heat often causes unwanted noise in the image. This shows up as graininess and blurry edges. Chip selection will be an important consideration to battle the effects of heat on image quality. The USB controller generates tremendous amounts of heat due to high-speed signaling and in serious cases where the USB controller is sitting next to the image sensor, the image actually shows bleeding and extreme degradation in quality on the side adjacent to the USB controller. Given the confined space within the typical notebook LCD bezel, designers need to design the board to separate the USB controller as far as possible from the image sensor, while choosing a USB controller with the lowest current consumption, which translates to lowest heat.

Simply put, dynamic range is how well the image sensor can handle extremely low light and extremely bright light. When evaluating image sensors, this is an important characteristic designers should benchmark, because it is very likely a lamp in the background can ruin the video capture. Figure 3 illustrates how an image sensor with poor dynamic range washes out on brighter parts of the picture (left) while an image sensor with good dynamic range maintains composure.

Figure 3: Better dynamic range, on the right, allows both detail in the car and outside the garage door to be visible.

Falling into the megapixel marketing trap results in severe degradation of low light performance and frame rate, which are limited by the typical user environment. Designers of notebook PCs can swing the ship back in the right direction by also focusing on mechanical constraints, architecture, resolution, optical format, frame rate, low light performance, USB throughput, heat and dynamic range when designing integrated notebook webcams. By focusing on these attributes which matter most, designers can improve image quality and end-user experience.

About the author
Shone Tran
is a product manager in Cypress Semiconductor's consumer and computation division.

Article Comments - In notebook webcams, less pixels mea...
*? You can enter [0] more charecters.
*Verify code:


Visit Asia Webinars to learn about the latest in technology and get practical design tips.

Back to Top