0
$\begingroup$

Signal processing is not my area of expertise so need help with implementing an OFDM system.

EDIT: The cabling and uncontrolled topology already exists and need a system that re-uses existing installation in large buildings

  1. System is for wired communication on a shielded twisted pair. The topology is a combination of daisy-chaining, star-connections at the nodes which further go onto more daisy-chaining. Essentially as bad as power lines but largely interference free because of shielding and differential signalling.
  2. Bandwidth is best kept as low as possible considering very long cable runs (half a kilometer) between nodes.
  3. Need to be able to send up to 256kbps.

Have been reading up on OFDM from an implementation point of view. I now understand a a bit about the use IFFT/FFT to inject and extract data out of the orthogonal carriers. Need for guard intervals, frame synchronization with pilot symbols, channel estimation with pilot careers etc. Things are still coming together in my head but I have a very basic understanding of OFDM. I have some questions:

  1. This being baseband is there a way of not having to use quadrature mixing to transmit the Real and Complex parts of the IFFT i.e. non-passband way of sending it?
  2. I am trying to keep the demodulation part as simple as possible. Anyway I can get away with not using channel estimation i.e. non-coherent demodulation with DPSK or something similar?
  3. Any pointers to fully implemented systems that I can dissect to learn more?
  4. Considering the low data rate, I'll probably end up doing this on a low-end DSP.

EDIT: There is only one differential pair in the cable. The bus will have 12 to 15 nodes on it with any-to-any communication.

$\endgroup$
  • $\begingroup$ You say something about the bit rate. How often does one node use that cable, or conversely, how many nodes try to use the bus on average, at once? I'm asking myself if a simple packetized system could theoretically work, or whether you're better of with "steady" links that schedule transmissions. $\endgroup$ – Marcus Müller Mar 26 '18 at 13:18
  • $\begingroup$ With current estimates, at any time no more than 3 nodes will be wanting to talk. When you say "steady" links - what do you have in mind please? $\endgroup$ – user183368 Mar 26 '18 at 13:21
  • $\begingroup$ you can either make something that allows nodes to send whenever they want and the channel is free (very much like WiFi does, usually), or you can just say "Here's one master node, it sets the time when each node is allowed to talk", from a network perspective. The first is easier to implement, usually, while the second makes more guarantees, and can, in larger networks, be significantly more efficient. $\endgroup$ – Marcus Müller Mar 26 '18 at 13:24
  • $\begingroup$ but "daisy chaining" sounds like the "non-end" nodes would have two transceivers – one "upstream" (in direction of the star "center") and one "downstream", which effectively means you could do point-to-point links. $\endgroup$ – Marcus Müller Mar 26 '18 at 13:26
  • $\begingroup$ I see what you mean, I am thinking CSMA/CA for simplicity and collision errors are handled by upper layers. Daisy chaining in the old-school sense where the cable is connected to a node and then continues to the next node. If you ever see old analog audio and security cables you'll get an idea of how uncontrolled it can be but it works for audio and very low data rate signaling $\endgroup$ – user183368 Mar 26 '18 at 13:31
1
$\begingroup$

You're describing DSL.

So, go and implement DSL! The common ADSL is an OFDM system.

This being baseband is there a way of not having to use quadrature mixing to transmit the Real and Complex parts of the IFFT i.e. non-passband way of sending it?

Yes, you can just make sure not to use one half of your carriers. That's mathematically not different from using a half as long DFT and and shifting it up by half-bandwidth.

I am trying to keep the demodulation part as simple as possible. Anyway I can get away with not using channel estimation i.e. non-coherent demodulation with DPSK or something similar?

I don't understand that question. OFDM is used exactly because you'd need an equalizer to use a broadband channel, and the channel estimator is OFDM, in essence.

Yes, there's OFDM systems that use a differential modulation on their subcarriers (namely, DAB/DAB+). The rationale, however, is usually a tradeoff between achievable rate and easy implementation: DAB has to deal with doppler spread, anyway, so it would require a significant amount of channel sensing if done non-differentially; if you're wasting much SNR on channel estimation, you can as well spend 3dB of BER curve and do differential modulation, especially if you're in the early 1990's and everything else would be computationally hard.

If you're doing fixed line communication, where the channel rarely changes, using a differential modulation is very inefficient compared to having a channel sensing preamble. You'd need to build a system with a higher bandwidth to achieve the same rate you could do with channel sensing, and that in a channel where you really want to avoid being forced to use higher frequencies.

Any pointers to fully implemented systems that I can dissect to learn more?

ADSL.

Considering the low data rate, I'll probably end up doing this on a low-end DSP.

... end up using existing DSL link technology instead of building this on your own.

I honestly don't think the modulation is the hard part here – the link and network layer becomes interesting. Twisted pair for a couple dozen meters sounds like you actually just want to use 10/100 Base-T. Modern network cards are surprisingly good at dealing with low-quality channels, and shielded twisted pair is actually a pretty nice channel (as opposed to what you claim, similarity to powerline).

If you really don't have the chance to implement half-duplex 10Base-T, you could probably simply go, physically, for simple serial RS-485; again, the problem is likely less a physical, but network layer design effort.

$\endgroup$
  • $\begingroup$ Thanks for the very detailed response. I edited my post to clarify a few things. There is only one pair in the cable. My first thoughts were to use 10Base2 but considering the long cable lengths(500 meters) and uncontrolled topology, it isn't ideal. OFDM fits the bill well but DSL in it's current form won't cut it as it's a point-to-point system and needs per-link training to be useful in a multi-node system? The closest bet will be (G3)PLC I believe, which I am still looking into but wondered if it would be easier to do the PHY implementation myself. I can tie it to ethernet MAC for the rest $\endgroup$ – user183368 Mar 26 '18 at 13:11
  • $\begingroup$ Do you think RS485 will work at those data rates on such long cable runs in a uncontrolled topology? My experience says otherwise but I could be wrong. $\endgroup$ – user183368 Mar 26 '18 at 13:12
  • $\begingroup$ When using RS-485, your link-to-link cost and complexity would be so low that instead of shared medium, you'd do single links with routing. Again, higher-layer problem. RS-485 is very robust, and over controlled, shielded cabling with strong drivers, half a km is possible, yet not easy. $\endgroup$ – Marcus Müller Mar 26 '18 at 13:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.