Visualizing Ethernet Speed

anthemaniac writes "In the blink of an eye, you can transfer files from one computer to another using Ethernet. And in the same amount of time, your eye sends signals to the brain. A study finds that images transferred to the brain and files across an Ethernet network take about the same amount of time." From the article: "The researchers calculate that the 100,000 ganglion cells in a guinea pig retina transmit roughly 875,000 bits of information per second. The human retina contains about 10 times more ganglion cells than that of guinea pigs, so it would transmit data at roughly 10 million bits per second, the researchers estimate. This is comparable to an Ethernet connection, which transmits information between computers at speeds of 10 million to 100 million bits per second."

Comments

So if I plug enough CAT5 cables into it...

[Siguy]

...I can use my guinea pig as a router?

Re:So if I plug enough CAT5 cables into it...

[evanbd]

Well, apparently you have to plug the cables into its eye sockets, and cramming more than one into each is probably hard. So more of a bridge than a router, I think...

Re:So if I plug enough CAT5 cables into it...

[Iron Condor]

Actually, if you wrap some duct tape around it you can jam a lot of stuff into guinea pigs...

Re:So if I plug enough CAT5 cables into it...

[ThJ]

I guess the human eye must use some hellishly good compression. I suspect the human eye of having more 'pixels' than good DV camera. PAL DV resolution is 720x288 at 50 fields/second or 720x576 at 25 frames/second. The human eye has 1 type of luma sensor, plus 3 types of chroma sensors. Humans have fewer sets of chroma sensors, let's assume a quarter of the number of luma sensors (similar to what YUV 4:2:2 assumes) If we assume a very poor human eye with this kind of resolution, we get:

720 * 576 = 414,720 luma
414,720 / 4 = 103,680 chroma
414,720 + 103,680 = 518,400 elements

Let us assume a poor human eye that can see only 25 frames/second in bright light conditions.

518,400 elements * 25 FPS = 12,960,000 elements/second

Add to this that the human eye probably has at least 8 "bits" of color resolution. How is an eye nerve that only transmits 8 "bits" a second going to convey even this relatively poor quality image?

Re:So if I plug enough CAT5 cables into it...

[Bonker]

Your numbers are probably a bit high.

Remember that humans don't see a pixel-per-pixel representation of the world. We see a tight spot of color and detail in the center of our retina (Fovea? Bio-types please correct me) surrounded by blurry shapes and lines. Around the edges, in peripheral we don't even see color, just luminance.

Proof? Take a bright LED lamp and move it into your peripheral vision. What color is it, not from memory, but just from looking at it?

The Fovea-area of the retina is more densely packed cells and blood vessels. It has more cones -- Chroma-type cells -- than rods.

This indistinct image is inverted and processed into a whole by the brain, which carefully processes different shapes, lines, movement, flickering, and what-not to produce what you THINK you see. The brain fills in any given pieces of the image that don't have enough detail, frequently from memory.

This is why optical illusions work. You deceive the biological mechanisms that process the image into producing bad data by giving them a skewed sample of the image.

Also, neural mechanisms are asynchronous and really can't be mesured in a k/s rate. You perceive a flicker of motion one second and then a spot of color the next. This is assembled into a ball that you turn to face-- to get a better image-- and then catch. Your brain has a lot of built-in firmware to do image manipulation, built you have to 'learn' the software necessary to do pattern matching and response over your lifetime.

You only get a few bits worth of information for the first few milliseconds that you're recognizing the ball, and then many megabytes worth the last few second.

Another thing... as sensitive and immersive as vision is, your ears probably have much, much more data input. They have vastly more dynamic range. Most people don't even notice themselves filling in visual information with audio information, but it does happen.

For example, you hear a person's voice, and you *think* you see their face.

Close your eyes when talking to someone, especially when in a group. Note how easy it is to visualize faces just from hearing voices.

I'm not denying that the brain has massive throughput from the senses, but you really shouldn't try to measure it in digital terms. It's all analogue.

Re:So if I plug enough CAT5 cables into it...

[Kadin2048]

As other people have pointed out, I think your estimates of the data that's actually being sent 'down the wire' from the eye to the brain is probably very high.

This came up in another discussion a while back, but I suspect that even an average digital camera with a good, wide-angle lens probably captures in a single frame more raw information than the human eye does from the same vantage point in a single glance.

You only think that your eye is a really good camera. In reality, it might be pretty bad -- I suspect that if you could watch the "raw feed" from a human eye on a TV screen, you might find it rather disappointing without all the postprocessing done in the brain's visual cortex. Only a small part of it near the center would be high resolution; only the center region would be color, and the periphery would be just good enough to detect movement, not much else.

I guess in a way you could call this "compression," but in reality it's more a credit to the brain and the way the 'receiver' is designed, to create the feeling of a huge, high-refresh-rate, 180-degree, full color, 3-D panorama, from not particularly impressive source imagery.

Basically, the human visual system makes up for the limitations of its cameras (the eyes) in post. In the synthetic machine world, we do not currently have the processing power nor the software necessary to do the kind of synthesization that the brain does, so instead we give the machines better 'eyes,' because building cameras is something we do know how to do.

A while ago, I heard someone who was involved in machine vision talk about something called the 'picket fence problem.' A person can pretty easily assemble a good idea of the scene on the other side of a picket fence, or a board with a few holes in it, by moving their head back and forth and then re-assembling the narrow-angle views into something more comprehensive, all in real time. I'm not sure whether machine-vision is there yet today (this was quite a while back), but it's a pretty non-trivial process, or so I was led to understand.

More interesting than the 'bitrate equivalence' of the optic nerve, would be some sort of estimation of the "processing power" done by an average person's visual cortex while doing some basic visual activity. I suspect that the result might be surprisingly high, maybe bordering on what would be supercomputer levels right now, solved using current methods.

Nice comparison

[Anonymous Coward]

This is comparable to an Ethernet connection, which transmits information between computers at speeds of 10 million to 100 million bits per second.

Yes, but we have better encoding.

Re:Nice comparison

[SnowZero]

It's not so much compression as having variable resolution. The center of the retina has a much higher resolution than the periphery. Try to look 20 degrees away from some words and try to read them (without looking back). It takes some practice to focus your attention a different place than where you are looking, but it's not too bad once you get the hang of it. You'll notice that there is very little detail indeed beyond about 5 degrees from the center of focus. The problem with TV and monitor displays is that they don't know where the viewer(s) are looking, so they need high resolution everywhere. Thus the eye can get away with a much lower overall information rate in comparison.

I do find that 10Mb to be a little low as an estimate however, since each cell can likely provide more than 10 bits/sec of information, especially when it can fire up to 1k times per second. There is almost surely less than 1000 bits/sec/cell though, so its somewhere between 100Mb and 1Gb overall -- still ethernet speeds.

There is a only a little bit of processing/compression going on at the eye itself; Mainly some "center surround" processing that you can roughtly compare to an edge sharpening filter in an image editing program. Most of the real processing goes on in the occipital lobe in the back of your head, such as the V1 layer which is essentially a 2D computer, with edge finding at various angles for every location in the eye. That feeds into V2 and V3, which do quite a bit more processing and are a little less well understood. One interesting thing is that it seems our visual systems split up into "what" and "where" pathways that independently process the identity of objects and their position. Some individuals with localized brain injuries can tell you they see an objecy (such as a stapler), but can't tell you where it is, and vice versa with an injury to the "what" pathway. Overall the human visual system is absolutely amazing, and we have a long way to go to catch up with it.

P.S. I know a lot about the human visual system because I have done a fair amount of computer vision research, and graduated with a double major in Cognitive Psychology (aka how the brain works).

Neuroscience != Computer Science

[KerberosKing]

I am not sure that thinking of signals from the eye to the brain work the same way as computer networks is very helpful. I don't think that there is the same sort of contention in a nervous system as there is in ethernet. Synapses as we understand them today do not appear to have any sort of collision detection. Neurons may have tens of thousands of other neurons that they are connected to in a many-to-one configuration and the whole process is analog, which is very different than ethernet frames. Also a single ganglion cell may send "10 million bits" of information, but the optic nerve is made of many such cells in parallel. I would not be surprised if our current estimates are wrong by at least an order of magnitude.

Re:Neuroscience != Computer Science

[davidsyes]

The eyes have it....

Well, I was all eyes for the article I partly read yesterday or early this am.

Well, even IF the eyes transmit like a network, the eye study is not apples and apples. More like oranges and mangoes.

How many libraries of eye sockets worth of information is that?

I mean, look at the size of the Guinea pig's eye. Of COURSE it transmits less energy. I mean more data to the brain. It's not as if it enlarges to accommodate more data. Hell, the human eye is probably 10 times LARGER without expanding. But, I suppose if the eyes DID expand when more data rate was demanded, such information overloads would lead to a whole new meaning of eye-socket-to-yah....

http://www.pimms-pages.co.uk/

And, Guinea Pigs aren't pigs of any sort. It's a terrible name to give something that a real pig could kill just by rolling over it.

http://www.oink.demon.co.uk/pets/guinea.htm

What *I* wanna know is how the eyes of bats comaare to the Guinea Pig and the Chupacabra. And to hell with human eyes. I'm talking about the MOVIE "Chupacabra".

Inaccurate blurb.

[pikine]

anthemaniac writes:

"In the blink of an eye, you can transfer files from one computer to another using Ethernet. And in the same amount of time, your eye sends signals to the brain. A study finds that images transferred to the brain and files across an Ethernet network take about the same amount of time."

The amount of time you transmit data over a network depends on round trip time and bandwidth product, which determines TCP window size that optimizes the send/ack of data packets. You also need to take collision into account.

The ganglion cells are probably more analogous to link transmitter. The measurement is on the amount of information generated by these cells per second. The proper conclusion is that you could probably use ethernet to connect the eyes and your brain, and the required bandwidth is supported.

Oh....

[ElephanTS]

I was wondering what that RJ45 socket on my head was for. My kids will probably be wireless.

but what about pirates with eye patches?!

[cryptonix]

Arr! they only get 10/half

Re:but what about pirates with eye patches?!

[ozbird]

Arr! they only get 10/half

With parroty?

Neurons make my head hurt

[BilZ0r]

The OP doesn't say that a single retinal cell transmits 10 million bits a second, but that the whole eye does. On top of that, while discussion of collition detection is pointless, thinking about the information a neuronal population can encode does have some merits. Although it's relatively pointless (at least now) to compare the eye to an ethernet, it has uses in comparing different neural populations.

The problem is that getting bitrates for neuronal populations is more of an art that a science. The sum total of information passed on by a neuron can not be computed simpley by it's spiking rate. Large numbers of parameters alter the actual chemical I/O relationship of a neuron. Resting membrane potential before spiking, whether it shows short term facilitation/depression etc...

Re:collision detection

[Anonymous Coward]

Just to pedantic here, "wireless Ethernet" does not use collision detection (CSMA/CD). it uses collision avoidance (CSMA/CA - i.e. 802.11), collision mitigation (CDMA - i.e. Navini, etc...), collision prevention (TDMA, polling, and their scheduling kin - i.e. Canopy, etc...), or it's simply FDD (modern expensive point-to-point, or old-school EoAMPS).

Even current wired Ethernet versions (1G, 10G) have dropped collision detection, opting to go full-duplex exclusively. Also shared cables can now carry multiple different signals without interferance, thanks to things like DWDM.

-l

So... Logically

[koterica]

My eithernet is the same speed as my eyes. My eyes can see my eithernet. My Eyes can see a duck. Therefor, if my eithernet weighs the same as a duck, its a witch!

AT&T

[richdun]

Oh great, now AT&T is going to charge me more to see certain things than others. Stupid eye neutrality.

(let's see how many pick up on the joke here...)

Um yeah, I dunno...

[DavidD_CA]

Here's how I look at it... the human eye has a "resolution" far greater than that which any monitor supports, and certainly greater than any streaming video I have ever seen.

Add to that the color depth of the human eye. Granted, not 16 M colors, but still pretty high.

The frame rate of the average human eye is somewhere around 40 fps, I believe. Again, faster than what most streaming videos offer.

Then double all that, 'cause we got two eyes.

I'm pretty sure the "bandwidth" between my eyes and brain is a little faster than even the best ethernet connection.. At least anything that I've seen demonstrated so far.

Re:Um yeah, I dunno...

[zizdodrian]

The frame rate of an eye isn't governed by the eye itself - given the transfer is analog, the frame rate is governed by what the visual cortex can handle. Which, if you have done any animation, is about 25 frames per second - a speed at which the human visual cortex cannot percieve the individual frames making up an animation.

I doubt the data transfer rate of any nerve is anywhere near the transfer speed of ethernet - I read an article once stating that human nerve tissue could transmit information at about 400 metres per second - something which ethernet stomps all over. The beauty of the brain is that it isn't restricted by raw transfer speed. What governs the speed at which the brain computes and calculates is its massively paralell connectivity. In fact, many have speculated that this connectivity is what conciousness itself arises from - it gives the brain a complexity far beyond the number of cells it actually contains.

The second thing to consider is that the brain is not limited by binary transfer - it can utilise chemicals and hormones, variable voltages, and timings to transfer information, not just on/off.

Re:Um yeah, I dunno...

[asuffield]

And you DO NOT have such resolution as your monitor.

Actually you do. However, you don't normally press your eyeball against the glass. At a distance of about .5m, which is normal, your eyes don't have sufficient resolution any more. If you move your head up close to the glass, you should be able to perceive the individual pixels. It's important to remember that there are three dimensions here. The expected viewing distance determines the necessary dpi of the viewing device.

But the data isn't "pixels" or anything....

[sbaker]

The numbers presented here are very misleading. You get the impression that your eyes are transferring video images as a bunch of pixels at the relatively slow speed of an Ethernet connection. But that's not true. Video processing starts right there in the retina and steadily changes the data from pixel-like date to edges, lines, shape to recognised objects to high level concepts that are conveniently tagged with memories, emotions and other relevent data.

At what point are we measuring the data? If the data that's actually being measured is something like "My Mom standing next to a table with a vase full of flowers on it" - then having 10 Mbits/sec is a heck of a lot of data. If it's raw video - then it's pathetically little.

We can estimate the bandwidth your eyes could theoretically produce if they were transmitting "raw video". We know that the retina has a resolution of around 5k x 5k "pixels" and we can see motion at around 60Hz and we have more dynamic range than we can display with 12 pixels each for Red, Green and Blue. So at the 'most raw', two eyes would require 5k x 5k x 60Hz x 2 x 12 x 3 bits per second. That's 108 Gbits/sec - which is vastly more than the 10Mbits to 100Mbits this article suggests. You can argue about the details of the numbers I used here - but we're looking at four orders of magnitude - so I have to be a LOT wrong!

So it's pretty certain that what they are measuring in TFA is some kind of condensed or summarized version of the visual data.

That being the case, it's pretty silly to be comparing "My Mom standing next to a table with a vase full of flowers on it" to a 640x480 JPEG file. It's simply not an 'apples and apples' comparison.

who cares about transfer speed?

[ltwally]

Who cares about the transfer speed. What I want to know is what kind of ping I'm getting.

Re:WOO HOOO 8th post

[homesteader]

Nothing to SEE here . . . 10mbit of bandwidth to see it with . . .