Fiber optics for dummies
October 6, 2009 11:35 AM
I have absolutely no comprehension whatsoever of how fiber optic cables can transmit information -- phone calls, emails, whatever -- by using light. Seriously, this makes no sense. Please explain it to me as if I'm a fifth-grader who makes decent grades in science class.
Listening to an NPR story about the Nobel winners this morning, I realized that I simply can't conceptualize how light can transmit my mother's voice on the telephone. And how did phone lines/internet work before they started using fiber optics?
Explain it simply?
Assume no prior knowledge or familiarity with acronyms.
Listening to an NPR story about the Nobel winners this morning, I realized that I simply can't conceptualize how light can transmit my mother's voice on the telephone. And how did phone lines/internet work before they started using fiber optics?
Explain it simply?
Assume no prior knowledge or familiarity with acronyms.
0s and 1s, dude, 0s and 1s. Light goes on, it's a 1. Light goes off, it's a 0. The devices on either end take care of turning the pulses back into meaningful information. It's really no different than how copper wire does the same thing with electricity. Light travels faster than conventional electrical signals, with less chance of interference or degradation, so it's a superior conductor of the 0s and 1s than copper. The fibers are much thinner than copper wires, so you can pack more of them into the same space that wires currently use, thus making them capable of providing more bandwidth.
I'm sure someone will show up soon with a more detailed answer, but I think this gets to the nut of what you're asking.
posted by briank at 11:43 AM on October 6, 2009
I'm sure someone will show up soon with a more detailed answer, but I think this gets to the nut of what you're asking.
posted by briank at 11:43 AM on October 6, 2009
I don't know how deeply you want to delve into the subject, but Understanding Fiber Optics by Jeff Hecht was a great resource for me. I come from a liberal arts background and ended up working for a fiber optics components company in 2001; I was able to work my way through (most of) that book on my own and finally feel like I understood what my company was manufacturing.
The current version is pretty pricey but it looks like Half.com has some older editions for cheap.
posted by something something at 11:43 AM on October 6, 2009
The current version is pretty pricey but it looks like Half.com has some older editions for cheap.
posted by something something at 11:43 AM on October 6, 2009
My understanding is that at base, it's just binary. Same as electrical impulses through a circuit. One end sends flickers of light, a photoreceptor on the other end receives and interprets.
posted by dirtynumbangelboy at 11:43 AM on October 6, 2009
posted by dirtynumbangelboy at 11:43 AM on October 6, 2009
Someone with extensive physics knowledge will doubtless come along and explain this better in a moment, but here's an easy way to start wrapping your head around it.
If you use a computer, you're familiar with the idea that lots of information can be transmitted as discrete bits. Everything you're reading on metafilter got into your computer as a series of ones and zeroes. Your computer then translates those patterns of bits into words, images, and sounds.
So for the simplest example of how light can be used to transmit information, think about a man with a flashlight, sending morse code. The light has two discrete states: on and off. These bits are then translated by the receiver into letters and words. This is essentially the same idea as using ones and zeros to create a digital representation of your mother's voice.
Real fiber optic technologies uses some clever tricks to increase the fidelity and bandwidth of the cables, but for a basic understanding, you can think of it as flashlights turning on and off and sending information down a very long, clear hallway.
posted by chrisamiller at 11:44 AM on October 6, 2009
If you use a computer, you're familiar with the idea that lots of information can be transmitted as discrete bits. Everything you're reading on metafilter got into your computer as a series of ones and zeroes. Your computer then translates those patterns of bits into words, images, and sounds.
So for the simplest example of how light can be used to transmit information, think about a man with a flashlight, sending morse code. The light has two discrete states: on and off. These bits are then translated by the receiver into letters and words. This is essentially the same idea as using ones and zeros to create a digital representation of your mother's voice.
Real fiber optic technologies uses some clever tricks to increase the fidelity and bandwidth of the cables, but for a basic understanding, you can think of it as flashlights turning on and off and sending information down a very long, clear hallway.
posted by chrisamiller at 11:44 AM on October 6, 2009
Okay, so think about morse code. You click the key, and a tappet at the other end clicks in sympathy. You are, in essence, sending pulses of electricity down the wire when you click the key, and those pulses cause the tappet at the other end to click. The timing and duration of your pulses have to be interpreted by a person at the other end to make any sense, and it's the timing and duration of your pulses that deliver the information.
Electronic devices work the same way, except that the pulses travel very, very, very fast, and so are sent and interpreted by computers. One computer clicks the key to send the information, the pulses go down the wire, and another computer interprets the timing and duration of the pulses to receive the information.
Now, fiber optics: replace the pulses of electricity with pulses of light. Instead of a key at the sending end, there's a light source to convert electrical pulses to light, and at the receiving end there's a light-sensitive receptor that converts light to electrical pulses. It's as simple as that.
posted by davejay at 11:46 AM on October 6, 2009
Electronic devices work the same way, except that the pulses travel very, very, very fast, and so are sent and interpreted by computers. One computer clicks the key to send the information, the pulses go down the wire, and another computer interprets the timing and duration of the pulses to receive the information.
Now, fiber optics: replace the pulses of electricity with pulses of light. Instead of a key at the sending end, there's a light source to convert electrical pulses to light, and at the receiving end there's a light-sensitive receptor that converts light to electrical pulses. It's as simple as that.
posted by davejay at 11:46 AM on October 6, 2009
Imagine we are on two different sides of a long bundle of 8 pipes each 6 inches in diameter. We are so far away that I can't see you and you can't see me. It's dark enough outside, though that when I hold a flashlight up to one of the pipes, you can see the light and tell which pipe it is coming from. So each pipe stands for a different bit in a byte. If the light is on, it is a 1, if not it is a 0.
A fiber optic cable is a bundle of clear fibers that are all aligned the same from one end to another, so if "fiber one" is lit up at one end, you can see the light at the other end. This is why it is so important to cut and join the cable properly.
posted by soelo at 11:49 AM on October 6, 2009
A fiber optic cable is a bundle of clear fibers that are all aligned the same from one end to another, so if "fiber one" is lit up at one end, you can see the light at the other end. This is why it is so important to cut and join the cable properly.
posted by soelo at 11:49 AM on October 6, 2009
Before fiber-optics, they used electricity. A microphone can convert pulses of air vibrations (sound) into pulses of electricity. At the other end, speakers convert pulses of electricty into air vibrations.
With fiber-optic technology, pulses of electricty are turned into pulses of light. This is done using things like LEDs (light emitting diodes) that you're probably familiar with on all kinds of devices with a power light. Similarly to speakers in the previous example, there are things that transform light into electric current: photodiodes.
The important breakthrough in fiber-optics is the fiber part. Basically, it's a material that reflects all light that bounces around the inside. Regular glass and plastic does this somewhat, if you've ever seen your reflection at night, looking out from your house window. The fiber-optic material does this for the light that is passing through it, making it bounce along without suffering too much degradation. You could think of it a little like a flexible tube with inward pointing mirrors all around. Using that tube and the LEDs and photodiodes on the ends of the tube, you can transmit bits at the speed of light.
posted by demiurge at 11:55 AM on October 6, 2009
With fiber-optic technology, pulses of electricty are turned into pulses of light. This is done using things like LEDs (light emitting diodes) that you're probably familiar with on all kinds of devices with a power light. Similarly to speakers in the previous example, there are things that transform light into electric current: photodiodes.
The important breakthrough in fiber-optics is the fiber part. Basically, it's a material that reflects all light that bounces around the inside. Regular glass and plastic does this somewhat, if you've ever seen your reflection at night, looking out from your house window. The fiber-optic material does this for the light that is passing through it, making it bounce along without suffering too much degradation. You could think of it a little like a flexible tube with inward pointing mirrors all around. Using that tube and the LEDs and photodiodes on the ends of the tube, you can transmit bits at the speed of light.
posted by demiurge at 11:55 AM on October 6, 2009
Imagine we are on two different sides of a long bundle of 8 pipes each 6 inches in diameter.
A fiber optic cable is a bundle of clear fibers that are all aligned the same from one end to another, so if "fiber one" is lit up at one end, you can see the light at the other end. This is why it is so important to cut and join the cable properly.
delving into this a bit further, sorry if this is a derail
we have eight pipes
what I don't understand is: how do I know that your signal in cable 1 at this exact nanosecond (call it moment 1) should be interpreted alongside or together with a signal in cable 2 at nanosecond 1.00001? There must be some lag sometimes between the cables? Or, imagine that one cable is just a little bit longer than the other? Shouldn't the communication degrade to nonsensical levels then?
And how many of those little cables you describe would be needed to have a decent phone conversation? Do you need 8 little cables? Or 64? Or would 1 do? I guess the question is related to: how fast must the cable be able to switch between on/off? Or is the bottleneck on the end of the sender and receiver?
posted by NekulturnY at 12:11 PM on October 6, 2009
A fiber optic cable is a bundle of clear fibers that are all aligned the same from one end to another, so if "fiber one" is lit up at one end, you can see the light at the other end. This is why it is so important to cut and join the cable properly.
delving into this a bit further, sorry if this is a derail
we have eight pipes
what I don't understand is: how do I know that your signal in cable 1 at this exact nanosecond (call it moment 1) should be interpreted alongside or together with a signal in cable 2 at nanosecond 1.00001? There must be some lag sometimes between the cables? Or, imagine that one cable is just a little bit longer than the other? Shouldn't the communication degrade to nonsensical levels then?
And how many of those little cables you describe would be needed to have a decent phone conversation? Do you need 8 little cables? Or 64? Or would 1 do? I guess the question is related to: how fast must the cable be able to switch between on/off? Or is the bottleneck on the end of the sender and receiver?
posted by NekulturnY at 12:11 PM on October 6, 2009
Remember how people aboard ships used to talk to other people aboard other ships by flashing lights and using Morse code?
It's just like that. Only the flashing lights are inside teeny tiny glass cables, and the code is binary, not Morse. And it happens really fucking fast.
posted by Cool Papa Bell at 12:11 PM on October 6, 2009
It's just like that. Only the flashing lights are inside teeny tiny glass cables, and the code is binary, not Morse. And it happens really fucking fast.
posted by Cool Papa Bell at 12:11 PM on October 6, 2009
Can someone explain how you can splice a fiber optic cable after Johnny Backhoe had cut one and still have light pass through the splice undistorted?
posted by JohnnyGunn at 12:12 PM on October 6, 2009
posted by JohnnyGunn at 12:12 PM on October 6, 2009
Can someone explain how you can splice a fiber optic cable after Johnny Backhoe had cut one and still have light pass through the splice undistorted?
The very short version: You cleanly cut the cable on either side of the damaged section and patch it correctly.
posted by Cool Papa Bell at 12:13 PM on October 6, 2009
The very short version: You cleanly cut the cable on either side of the damaged section and patch it correctly.
posted by Cool Papa Bell at 12:13 PM on October 6, 2009
During the splice process, the splicers line up each fiber-optic tube and fuse them back together. However, each splice point also introduces some degradation on the cable, so that particular cable will be a less efficient transmitter of information after each cut.
posted by hilaritas at 12:18 PM on October 6, 2009
posted by hilaritas at 12:18 PM on October 6, 2009
I think we've all seen graphs of sound with its peaks and valleys. That's called a continuous analog signal. And if we have an analog connection (say, microphone -> amplifier -> speaker) then it's just basic electronics to reproduce it at the other end. Hopefully you understand this concept.
But the modern world is digital, which means ones and zeros in tidy little packages. How do we get that continuous analog signal into a discrete digital one? Simple!
The first step is to take that analog signal and sample it at regular intervals. If you sample frequently enough, you can exactly reconstruct the original signal from the samples[1]. For demonstration purposes, let's say you sample a signal at 1 second intervals (we actually do it thousands of times per second, but this is just a thought exercise). Now we have a sequence of values like this:
{T0=1.2, T1=2.0, T2=2.2, T3=1.0, T4=0.4, T5=0.0}
Where Tx is the sample time, and the value is the height of the curve at that moment. This is a discrete analog signal, which is halfway there! It's discrete because we've broken the continuous curve into separate points. And it's analog because we're measuring the values in a fractional way with decimal points.
To make it digital, we assign each value level a number between 0 and 255. This is called quantization. So for our example, let's say 0 = 0, 1 = 0.2, 2 = 0.4, etc. That makes our signal now look like this (dividing each of the analog values by 0.2):
{T0=6, T1=10, T2=11, T3=5, T4=2, T5=0}
Now we finally have a discrete digital signal and we're ready to transmit!
Let's say I'm making a phone call from Chicago to London. My voice is going to be converted to a voltage curve by the microphone, then sampled and quantized into a digital signal by the phone. But how to get that digital signal to London? One way would be with a dedicated circuit. When I dial the number, the phone company reserves an exclusive path for me from Chicago to London. No one else gets to use that underwater fiber while I'm on the phone. From there, it's straight forward to just transmit the digital signal over the dedicated line.
But that's horribly inefficient. Fiber (and copper even) can transmit data much, much faster than I can talk. By dedicating a path for my exclusive use, we're wasting a lot of bandwidth. So the telephone companies (and everyone else) uses what's called a packet switched network.
My digital signal is broken up into packets of data. For our example, let's suppose each packet contains two samples:
{T0=6, T1=10}, {T2=11, T3=5}, {T4=2, T5=0}
Then each packet is assign a destination (the phone number I'm calling) and a sequence number[2]. So they might look like this:
{D=44-20-8433-2000, S=0, T0=6, T1=10},
{D=44-20-8433-2000, S=1, T2=11, T3=5},
{D=44-20-8433-2000, S=2, T4=2, T5=0}
Finally, each packet is routed separately to the destination. This lets the company schedule packets from everyone's phone calls together, over the same cable. My packets are interleaved with other peoples. And in fact, the first packet might leave from NYC and the second two from DC. All depends on what routes are available when the packets are scheduled.
Once the packets arrive at the destination, the sequence numbers are used to re-assemble them into the original signal. And from there, getting the analog signal back is as simple as reversing the process we went through above.
As for the physical transmitting of packets, other people have that nicely covered. All you need to know is that you can represent any integer as a sequence of ones and zeros. From there, it's just turning the light on and off really fast.
[1] It's the Nyquist–Shannon sampling theorem
[2] Packets contain more information, but we're just illustrating here.
posted by sbutler at 12:34 PM on October 6, 2009
But the modern world is digital, which means ones and zeros in tidy little packages. How do we get that continuous analog signal into a discrete digital one? Simple!
The first step is to take that analog signal and sample it at regular intervals. If you sample frequently enough, you can exactly reconstruct the original signal from the samples[1]. For demonstration purposes, let's say you sample a signal at 1 second intervals (we actually do it thousands of times per second, but this is just a thought exercise). Now we have a sequence of values like this:
{T0=1.2, T1=2.0, T2=2.2, T3=1.0, T4=0.4, T5=0.0}
Where Tx is the sample time, and the value is the height of the curve at that moment. This is a discrete analog signal, which is halfway there! It's discrete because we've broken the continuous curve into separate points. And it's analog because we're measuring the values in a fractional way with decimal points.
To make it digital, we assign each value level a number between 0 and 255. This is called quantization. So for our example, let's say 0 = 0, 1 = 0.2, 2 = 0.4, etc. That makes our signal now look like this (dividing each of the analog values by 0.2):
{T0=6, T1=10, T2=11, T3=5, T4=2, T5=0}
Now we finally have a discrete digital signal and we're ready to transmit!
Let's say I'm making a phone call from Chicago to London. My voice is going to be converted to a voltage curve by the microphone, then sampled and quantized into a digital signal by the phone. But how to get that digital signal to London? One way would be with a dedicated circuit. When I dial the number, the phone company reserves an exclusive path for me from Chicago to London. No one else gets to use that underwater fiber while I'm on the phone. From there, it's straight forward to just transmit the digital signal over the dedicated line.
But that's horribly inefficient. Fiber (and copper even) can transmit data much, much faster than I can talk. By dedicating a path for my exclusive use, we're wasting a lot of bandwidth. So the telephone companies (and everyone else) uses what's called a packet switched network.
My digital signal is broken up into packets of data. For our example, let's suppose each packet contains two samples:
{T0=6, T1=10}, {T2=11, T3=5}, {T4=2, T5=0}
Then each packet is assign a destination (the phone number I'm calling) and a sequence number[2]. So they might look like this:
{D=44-20-8433-2000, S=0, T0=6, T1=10},
{D=44-20-8433-2000, S=1, T2=11, T3=5},
{D=44-20-8433-2000, S=2, T4=2, T5=0}
Finally, each packet is routed separately to the destination. This lets the company schedule packets from everyone's phone calls together, over the same cable. My packets are interleaved with other peoples. And in fact, the first packet might leave from NYC and the second two from DC. All depends on what routes are available when the packets are scheduled.
Once the packets arrive at the destination, the sequence numbers are used to re-assemble them into the original signal. And from there, getting the analog signal back is as simple as reversing the process we went through above.
As for the physical transmitting of packets, other people have that nicely covered. All you need to know is that you can represent any integer as a sequence of ones and zeros. From there, it's just turning the light on and off really fast.
[1] It's the Nyquist–Shannon sampling theorem
[2] Packets contain more information, but we're just illustrating here.
posted by sbutler at 12:34 PM on October 6, 2009
Another concept if you are interested: total internal reflection Check out the Picture, it makes it much easier to understand part of the physics. This phenomenon helps reduce the amount of light lost during transmission, increasing the distance over which you can send a signal before you need to repeat/amplify it again.
Also, multiplexing allows multiple signals to be sent over the same physical cable (or individual wire in a cable). This way you don't have to wait for one piece of information to finish before starting another, you can send them both at the same time. With fiber optics, you can use different wavelengths (colors for example) on the same cable, with one piece of equipment on each end of the cable that only cares about "blue" light signals or "red" light signals. In reality, though, the wavelength differences are small, and the naked eye cannot tell them apart.
posted by kenbennedy at 12:54 PM on October 6, 2009
Also, multiplexing allows multiple signals to be sent over the same physical cable (or individual wire in a cable). This way you don't have to wait for one piece of information to finish before starting another, you can send them both at the same time. With fiber optics, you can use different wavelengths (colors for example) on the same cable, with one piece of equipment on each end of the cable that only cares about "blue" light signals or "red" light signals. In reality, though, the wavelength differences are small, and the naked eye cannot tell them apart.
posted by kenbennedy at 12:54 PM on October 6, 2009
Great answers, sbutler and kenbennedy. I think I even understand it.
Another one? Why 255 values? I see that number a lot when it comes to networks. Is there a special magic to it? I thought bits and bytes and stuff went in blocks of eights and fours (8, 32, 64, 128 etc.)
posted by NekulturnY at 12:57 PM on October 6, 2009
Another one? Why 255 values? I see that number a lot when it comes to networks. Is there a special magic to it? I thought bits and bytes and stuff went in blocks of eights and fours (8, 32, 64, 128 etc.)
posted by NekulturnY at 12:57 PM on October 6, 2009
"The important breakthrough in fiber-optics is the fiber part. Basically, it's a material that reflects all light that bounces around the inside. Regular glass and plastic does this somewhat, if you've ever seen your reflection at night, looking out from your house window. The fiber-optic material does this for the light that is passing through it, making it bounce along without suffering too much degradation. You could think of it a little like a flexible tube with inward pointing mirrors all around. Using that tube and the LEDs and photodiodes on the ends of the tube, you can transmit bits at the speed of light."
All materials do this to some extent (obviously materials that do not transmit light do this to the least extent). The degree to which it is effective or usable depends on the refractive index of the "carrier" material and "environment" material, which in the case of fiber optics would be the fiber and its jacket. You can get the same effect with a clear 2 liter bottle and water. Use a laser pointer and change the angle at which you shine it in, and you may be able to get it to bounce around inside. Also check out this pic, of total internal reflection you may have experienced yourself : wikipedia pic
Every combination of 2 materials will have a different angle at which total internal reflection occurs. If you miss that angle, you lose information.
posted by kenbennedy at 1:01 PM on October 6, 2009
All materials do this to some extent (obviously materials that do not transmit light do this to the least extent). The degree to which it is effective or usable depends on the refractive index of the "carrier" material and "environment" material, which in the case of fiber optics would be the fiber and its jacket. You can get the same effect with a clear 2 liter bottle and water. Use a laser pointer and change the angle at which you shine it in, and you may be able to get it to bounce around inside. Also check out this pic, of total internal reflection you may have experienced yourself : wikipedia pic
Every combination of 2 materials will have a different angle at which total internal reflection occurs. If you miss that angle, you lose information.
posted by kenbennedy at 1:01 PM on October 6, 2009
Your Mom -> Sound Waves -> Microphone -> Diaphram -> Analog Pulse -> Computer -> Digitized Pulse -> Fiber Optic -> Digitized Pulse -> Computer -> Electric Pulse -> Diaphram -> Speaker -> Sound Waves -> Your Ear
posted by spoons at 1:03 PM on October 6, 2009
posted by spoons at 1:03 PM on October 6, 2009
255: Which numbers can you represent using 1 digit? 0..9. Which ones using 2 digits? 0..99. 3 digits: 0..999. See the pattern? So using 8 bits, you can represent the numbers 0..255. (In binary, 255 is 11111111 and 256 is 100000000.)
posted by phliar at 1:04 PM on October 6, 2009
posted by phliar at 1:04 PM on October 6, 2009
Why 255 values?
The range 0 to 255 gives 256, or 2^8, integer values.
posted by utsutsu at 1:06 PM on October 6, 2009
The range 0 to 255 gives 256, or 2^8, integer values.
posted by utsutsu at 1:06 PM on October 6, 2009
255: Which numbers can you represent using 1 digit? 0..9. Which ones using 2 digits? 0..99. 3 digits: 0..999. See the pattern? So using 8 bits, you can represent the numbers 0..255. (In binary, 255 is 11111111 and 256 is 100000000.)
cool, thanks!
posted by NekulturnY at 1:07 PM on October 6, 2009
cool, thanks!
posted by NekulturnY at 1:07 PM on October 6, 2009
I just chose 255 because it's a nice, round binary number. Remember, this is just an example :)
However, different quantization schemes use different levels. I believe CD's use 16bit values, which means 0 to 65,535.
posted by sbutler at 1:16 PM on October 6, 2009
However, different quantization schemes use different levels. I believe CD's use 16bit values, which means 0 to 65,535.
posted by sbutler at 1:16 PM on October 6, 2009
demiurge: "The fiber-optic material does this for the light that is passing through it, making it bounce along without suffering too much degradation."
It also turns out that if you bend fiber optics too hard, they will "leak" some light. Not all of it, but enough to snoop what's going out across it. Communication devices generally assume some amount of lost signal, so they will cope with this seamlessly. If you bend it too hard, you will break the light bouncing geometry, but it was demonstrated that leakage happens before that.
posted by pwnguin at 1:27 PM on October 6, 2009
It also turns out that if you bend fiber optics too hard, they will "leak" some light. Not all of it, but enough to snoop what's going out across it. Communication devices generally assume some amount of lost signal, so they will cope with this seamlessly. If you bend it too hard, you will break the light bouncing geometry, but it was demonstrated that leakage happens before that.
posted by pwnguin at 1:27 PM on October 6, 2009
Oh, one other thing: the photoelectric effect is what lets electronic computers read light. Some special silicon lets electricity flow as photons hit it. More photons == more electric flow. Once you add that into the mix, it's not much different than normal signal processing we do with copper lines, with quantitization and multiplexing and spread spectrum etc.
posted by pwnguin at 1:36 PM on October 6, 2009
posted by pwnguin at 1:36 PM on October 6, 2009
NekulturnY, you asked:
what I don't understand is: how do I know that your signal in cable 1 at this exact nanosecond (call it moment 1) should be interpreted alongside or together with a signal in cable 2 at nanosecond 1.00001? There must be some lag sometimes between the cables? Or, imagine that one cable is just a little bit longer than the other? Shouldn't the communication degrade to nonsensical levels then?
This is an important issue in digital communications, and there are various ways of solving it. Part of the solution for a lot of long-distance transmission is that you don't use a parallel interface, where you read the signals across multiple channels at the same time, which is what soelo described, you use a serial interface where you read subsequent bits from a single channel and chop them up into predefined sizes to get each chunk of data.
There is still the problem of figuring out where one bit ends and the other begins. I think the solution generally used is to have highly accurate "clocks" at each end and agreement on how many clock ticks each bit lasts. Those clocks are kept in sync using signals sent over the communications line. At the simplest, you might say that each '1' is transmitted as a dim light, each '0' as no light, and that each bit will be interleaved with a very bright light. Then you just look at your signal, find the peaks, and then look at the intensity between the peaks to find the value of each bit. In practice this approach ends up squandering a lot of potential bandwidth on carrying timing information, so, one way or another, more bits get stuffed through between timing pulses, with accurate clocks providing confidence in separating each bit.
Actually, there are all sorts of things that can be done to stuff more data down a channel more accurately. For one thing, they make use of shades of grey. Rather than just sending one bit in a cycle by turning the light on or off, a pulse may have a range of intensities, and encode a number of bits.
posted by Good Brain at 1:47 PM on October 6, 2009
what I don't understand is: how do I know that your signal in cable 1 at this exact nanosecond (call it moment 1) should be interpreted alongside or together with a signal in cable 2 at nanosecond 1.00001? There must be some lag sometimes between the cables? Or, imagine that one cable is just a little bit longer than the other? Shouldn't the communication degrade to nonsensical levels then?
This is an important issue in digital communications, and there are various ways of solving it. Part of the solution for a lot of long-distance transmission is that you don't use a parallel interface, where you read the signals across multiple channels at the same time, which is what soelo described, you use a serial interface where you read subsequent bits from a single channel and chop them up into predefined sizes to get each chunk of data.
There is still the problem of figuring out where one bit ends and the other begins. I think the solution generally used is to have highly accurate "clocks" at each end and agreement on how many clock ticks each bit lasts. Those clocks are kept in sync using signals sent over the communications line. At the simplest, you might say that each '1' is transmitted as a dim light, each '0' as no light, and that each bit will be interleaved with a very bright light. Then you just look at your signal, find the peaks, and then look at the intensity between the peaks to find the value of each bit. In practice this approach ends up squandering a lot of potential bandwidth on carrying timing information, so, one way or another, more bits get stuffed through between timing pulses, with accurate clocks providing confidence in separating each bit.
Actually, there are all sorts of things that can be done to stuff more data down a channel more accurately. For one thing, they make use of shades of grey. Rather than just sending one bit in a cycle by turning the light on or off, a pulse may have a range of intensities, and encode a number of bits.
posted by Good Brain at 1:47 PM on October 6, 2009
Thanks, Good Brain.
posted by NekulturnY at 2:41 PM on October 6, 2009
posted by NekulturnY at 2:41 PM on October 6, 2009
Read Mother Earth, Motherboard by Neal Stephenson. A fantastic history lesson in undersea communications.
posted by PenDevil at 2:54 PM on October 6, 2009
posted by PenDevil at 2:54 PM on October 6, 2009
Use the printing link rather. The sucker is 56 pages long.
posted by PenDevil at 2:55 PM on October 6, 2009
posted by PenDevil at 2:55 PM on October 6, 2009
Good Brain: "I think the solution generally used is to have highly accurate "clocks" at each end and agreement on how many clock ticks each bit lasts"
Actually, what happens in my experience is you use a line code to make clockless signals. In essence, you declare some minimum resolution timer for a quasi-bit, and then use the transition direction to indicate a 0 or a 1. The gain here is that the clock is reset after every bit so drift is negated.
posted by pwnguin at 10:24 PM on October 6, 2009
Actually, what happens in my experience is you use a line code to make clockless signals. In essence, you declare some minimum resolution timer for a quasi-bit, and then use the transition direction to indicate a 0 or a 1. The gain here is that the clock is reset after every bit so drift is negated.
posted by pwnguin at 10:24 PM on October 6, 2009
Whoops, thats exactly what you describe. I misread it as "have highly accurate 'clocks' at each end in agreement." Always helps to read the whole thing.
posted by pwnguin at 10:27 PM on October 6, 2009
posted by pwnguin at 10:27 PM on October 6, 2009
NekulturnY, most of the circuits I've designed in the past couple years relate to lining up bits that come across a multi-conductor cable (fibre optic or copper).
The cable itself, the connector, the printed circuit board and the ASIC all add their own skew components so you're correct, in general bits that are sent down the wire at the same time don't arrive on the other side at the same time.
To compensate that deskewing circuits are built. Before working with real data, training sequences are sent down the wires. These sequences contain code words that you can think of as markers. If the markers are spaced in time wider than the potential skew between the transmitter and receiver it's possible to null out the skew.
Suppose as an example we have a pair of cables and on each of the cables we transmit words of information by a sequence of 1s and 0s. We reserve one sequence as a marker for lane alignment. We know there's less than 5 nanoseconds of skew in the cables, PCB, etc. (because we specify that, and buy cables that meet that specification and design our PCB so that it's met etc). Our system also has a clock that toggles every 0.5 ns.
During training we force the markers to be written into a memory at the 0th address, and there's one memory for each cable. We time the markers and size the memories so that each time they occur we've wrapped around the memory such that we're at address 0 again. Since skew doesn't vary very much very quickly. This means that we've now aligned the data in memory.
The second part is we need to know how much skew there is so we don't read from memory before a full word is written. To do this when we detect the first marker in whatever cable it happens to occur in we start counting. One clock, two clock, three clock etc. When we detect the second marker we stop, say at 7 clocks, we stop counting.
We now know that when we're writing to an address we can only read from an address at least 7 addresses less than it. Once we're trained we go into mission mode.
This is simplified but captures most of the details.
posted by substrate at 1:49 PM on October 7, 2009
The cable itself, the connector, the printed circuit board and the ASIC all add their own skew components so you're correct, in general bits that are sent down the wire at the same time don't arrive on the other side at the same time.
To compensate that deskewing circuits are built. Before working with real data, training sequences are sent down the wires. These sequences contain code words that you can think of as markers. If the markers are spaced in time wider than the potential skew between the transmitter and receiver it's possible to null out the skew.
Suppose as an example we have a pair of cables and on each of the cables we transmit words of information by a sequence of 1s and 0s. We reserve one sequence as a marker for lane alignment. We know there's less than 5 nanoseconds of skew in the cables, PCB, etc. (because we specify that, and buy cables that meet that specification and design our PCB so that it's met etc). Our system also has a clock that toggles every 0.5 ns.
During training we force the markers to be written into a memory at the 0th address, and there's one memory for each cable. We time the markers and size the memories so that each time they occur we've wrapped around the memory such that we're at address 0 again. Since skew doesn't vary very much very quickly. This means that we've now aligned the data in memory.
The second part is we need to know how much skew there is so we don't read from memory before a full word is written. To do this when we detect the first marker in whatever cable it happens to occur in we start counting. One clock, two clock, three clock etc. When we detect the second marker we stop, say at 7 clocks, we stop counting.
We now know that when we're writing to an address we can only read from an address at least 7 addresses less than it. Once we're trained we go into mission mode.
This is simplified but captures most of the details.
posted by substrate at 1:49 PM on October 7, 2009
The other approach, of course, is to simply calculate the maximum possible worst case skew, multiply by a safety factor of five or ten, and call it good. Often in this case you'll have one line whose purpose is simply to carry a pulse indicating that the other lines currently contain good data. Set up the data on some lines, wait a bit, pulse the clock/strobe line, wait some more without changing the data, then move on to the next word. Obviously this doesn't pump as much data through the system as more sophisticated techniques could, but in a lot of cases it's all you need. If you're only going a few inches from one side of a logic board to the other, for example, then the "waits" can be quite short.
Synchronization and clock-recovery is a large, sprawling topic; there are a zillion techniques out there, some complicated, some simple, some used everywhere, some really specialized. For long distances, they usually involve interleaving some known, fixed pattern in with the real data; the receiver knows about this and adjusts its timing so that the fixed pattern comes out right, and the rest of the data should come out right as well.
posted by hattifattener at 3:02 AM on October 8, 2009
Synchronization and clock-recovery is a large, sprawling topic; there are a zillion techniques out there, some complicated, some simple, some used everywhere, some really specialized. For long distances, they usually involve interleaving some known, fixed pattern in with the real data; the receiver knows about this and adjusts its timing so that the fixed pattern comes out right, and the rest of the data should come out right as well.
posted by hattifattener at 3:02 AM on October 8, 2009
Clock synchronization on both transmit and receive ends is a lot like how a band of musicians gets synced up on the beat. In a symphony orchestra you have the conductor's baton which everyone watches to determine the beat and tempo. But having a conductor in a small band is not practical because it means you have one member that does nothing except set the beat. In communications this is also inefficient because it means you need one channel to send the clock or beat and another to send the data or notes.
So for most high speed communications they use a scheme called self-clocking in which the clock is embedded in the data on the same channel. This is like a small band without a conductor. The band leader starts off a song with a count of four quarter note beats and tapping his foot -- one, two, three, four. By the fourth note everyone is synchronized and on the beat and tapping their own foot. In the same way, each information packet (or song) on the fiber begins with a series of sync pulses that allow the receiver to synchronize with the beat before the data (song) starts.
Everyone in the band is now tapping their own foot and they don't even have to look at the band leader anymore. They can close their eyes since they have their own tapping foot to keep the time. Even if there are whole notes, they can measure out the beginning and end of each four-beat note because they have their own internal tapping foot as a clock. The fiber link works the same way. Once they get synchronized at the beginning of the packet, the receiver can pick off the bits in the rest of the packet using its synchronized clock.
But if the song were to consist of nothing but whole notes for a long time, the band members might begin to drift slightly apart because their aren't enough quarter notes to keep everyone synchronized on the beat. Fortunately most music isn't all whole notes. There are always quarter notes here and there that allow the band members to again get precisely on the beat and fine tune their tapping foot through the end of the song. Likewise for communications, the data packets are always designed to have some portion of quarter notes in the packet so that the receiver can stay on the beat to the end of the packet (song).
Using this self-clocking method, the clock or beat is embedded in the data or song itself. You don't need a separate clock or conductor to keep the time.
posted by JackFlash at 10:50 AM on October 10, 2009
So for most high speed communications they use a scheme called self-clocking in which the clock is embedded in the data on the same channel. This is like a small band without a conductor. The band leader starts off a song with a count of four quarter note beats and tapping his foot -- one, two, three, four. By the fourth note everyone is synchronized and on the beat and tapping their own foot. In the same way, each information packet (or song) on the fiber begins with a series of sync pulses that allow the receiver to synchronize with the beat before the data (song) starts.
Everyone in the band is now tapping their own foot and they don't even have to look at the band leader anymore. They can close their eyes since they have their own tapping foot to keep the time. Even if there are whole notes, they can measure out the beginning and end of each four-beat note because they have their own internal tapping foot as a clock. The fiber link works the same way. Once they get synchronized at the beginning of the packet, the receiver can pick off the bits in the rest of the packet using its synchronized clock.
But if the song were to consist of nothing but whole notes for a long time, the band members might begin to drift slightly apart because their aren't enough quarter notes to keep everyone synchronized on the beat. Fortunately most music isn't all whole notes. There are always quarter notes here and there that allow the band members to again get precisely on the beat and fine tune their tapping foot through the end of the song. Likewise for communications, the data packets are always designed to have some portion of quarter notes in the packet so that the receiver can stay on the beat to the end of the packet (song).
Using this self-clocking method, the clock or beat is embedded in the data or song itself. You don't need a separate clock or conductor to keep the time.
posted by JackFlash at 10:50 AM on October 10, 2009
So the telephone companies (and everyone else) uses what's called a packet switched network...
I'm curious about this. Up to about fifteen years ago, the telephone network was heavily circuit switched. What's the penetration of packet switching these days? Is it correct that it's 100%?
Is Signaling System 7 still used for telephone call set up? If so, then even if the voice data is packet switched, it's unlikely that the phone number would be contained in the packet as given in the example.
posted by storybored at 11:48 AM on October 11, 2009
I'm curious about this. Up to about fifteen years ago, the telephone network was heavily circuit switched. What's the penetration of packet switching these days? Is it correct that it's 100%?
Is Signaling System 7 still used for telephone call set up? If so, then even if the voice data is packet switched, it's unlikely that the phone number would be contained in the packet as given in the example.
posted by storybored at 11:48 AM on October 11, 2009
This reminds me of a (possibly apocryphal) quote from Albert Einstein, who, when asked to explain how wireless (radio) works, said:
“The wireless telegraph is not difficult to understand. The ordinary telegraph is like a very long cat. You pull the tail in New York, and it meows in Los Angeles. The wireless is the same, only without the cat.”
So, you might say, fiber optics work the same way, but with an optical cat....
posted by crazy_yeti at 11:07 AM on October 20, 2009
“The wireless telegraph is not difficult to understand. The ordinary telegraph is like a very long cat. You pull the tail in New York, and it meows in Los Angeles. The wireless is the same, only without the cat.”
So, you might say, fiber optics work the same way, but with an optical cat....
posted by crazy_yeti at 11:07 AM on October 20, 2009
"The wireless is the same, only without the cat.”
so let me get this straight; does peta endorse wireless communication and denounce all wired communication?
posted by infinite intimation at 2:24 PM on October 24, 2009
so let me get this straight; does peta endorse wireless communication and denounce all wired communication?
posted by infinite intimation at 2:24 PM on October 24, 2009
This thread is closed to new comments.
posted by doteatop at 11:42 AM on October 6, 2009