--Towards Collision Detection in Wireless Networks

Collision occurs when 2 hosts cannot hear each other.  This is different from ethernet, where we have CSMA/CD; which can allow for more efficient use of network resources.  But if we can somehow detect collisions and communicate them over a side-channel, we should be able to get something similar in wireless.

We need to be able to detect collisions and notify neighbors in realtime, but the power of our self-signal can make it difficult to identify colliding signals.  But in this case, we don't need to actually try to decode a signal - if we replace the idea of decoding  and just attempt to find a correlation with the unique preamble from another node, then we can still manage to detect colliding signals.

In experimental tests with overlapping nodes, the CSMA/CN technique manages to abort terminal cases earlier then traditional wireless methods, and in highly congested networks, we see up to a 50% throughput increase.

Future work:

Look at interactions with multiple interfering nodes and at how notification can cause interference.

Questions:

Q. What is the effect on correlation if a single node is sending to both a nearby and a far away node?

We haven't looked at broadcast receiver correlation yet - but this can possibly be addressed by having a shared signature for broadcast addresses.

Q.  How do you compute signatures, since it seems a requirement that they uniquely identify each transmitting pair?  Code assignment seems like a hard problem.

We're looking into the assignment of signatures now; current work has manually assigned signatures.

Q. At what range does this work - it seems like self-signal would dominate?

Decoding is hard; but correlation is easier.

(Some discussion about frequency offsets, and self-suppression).

Q.  What is the difference between this and falling back to a slower rate?

The speaker offered a comparison with BPMA

(More discussion regarding elimination of self signal)

-- Retransmission != Repeat: Simple Retransmission Permutation Can Resolve Overlapping Channel Collisions

We can permute the bits assigned to our subcarriers on retransmission to resolve overlapping element collisions.

Normal 802.11 channels have 4 subcarrier groups - bits are assigned to the subcarriers.  If we cycle bits across our subcarriers, then we can reconstruct packets from the non-matching collisions.  This can be generalized to other channel types.

Q. How do you know what to subtract out if you haven't acquired enough channels to decode the packet?

You only decode out the subchannels that you have acquired enough information on.  Zig/zag + remapping allows you to see more.

-- One-size fits all wifi videos

Wireless video is a increasingly large space - mobile TV and digital broadcast TV are rapidly growing.  All involve some kind of multicast and mobility on the part of the receiver; but this mobility has a price - the bit rate available to the receiver varies on location, and properly encoding to multiple bit rates is hard.

Single-bit errors in traditional codecs like MPEG can radically alter the decoder state and cause poor quality imagery.

(Overview of image encoding via DCT)

Rather then computing a frequency table for an image as reals, and then encoding from binary to reals, we can directly transmit the frequency weights from our DCT onto the physical layer, reweighting as appropriate to avoid using too much power and to provide reasonable error characteristics.  A small error in the physical channel results in a small error in our decoded video.

(Presentation of experimental results using noise information collected from real receivers).

Q. How do you exploit inter-frame encoding?  This is fundamental to things like MPEG.

We actual do encode blocks of frames in our code; the demonstration is using this technique.

Q. You mentioned losing signal could result in losing pixels from a normal encoder, but it seems like you could also lose whole frames if say, your frame header is damaged.

The slide was a simplification - in reality, it is even harder to correctly handle poor channels.

Q. What are the tradeoffs?  What do you give up when switching to direct floating-point encoding?

Nothing really.  This encoding maps better to a wireless channel; it wouldn't be better if we were to write it to disk, but for the wireless environment we're looking at, it maps better.

Q.  How expensive is DCT against the whole image - JPEG does blocking to avoid computing the whole frame at once.

We found the runtime is very similar for our test cases using blocked and whole frame DCT.

Q.  Doesn't interframe handling make your scheme more complicated?

We handle it by encoding multiple frames as a block (computing a 3d frequency table).