SDSL services are normally advertised and sold at speeds that are nice round numbers, i.e., multiples of 64 kbps traditional in the telecom world. But surprisingly the actual bit rate at which the bitpump is operated is very often an odd-looking number that is not a nice round multiple of 64 kbps, even with SDSL flavors A and B which involve no framing overhead.

It appears that the bitpump's analog line interface (both the on-chip analog circuitry and the external hybrid) works better at some data rates than others. It is understandable that this analog circuitry is sensitive to the frequency spectrum of the signal, which is in turn directly related to the symbol rate. Apparently this sensitivity was even greater in the earlier Bt8970 bitpump than in the RS8973. The former was designed to operate primarily at HDSL bit rates of 784 and 1168 kbps, whereas the latter was designed with multirate SDSL applications in mind.

The RS8973 bitpump's built-in clock synthesiser can produce any data rate that is a multiple of 8 kbps and the device is advertised as capable of operation from 144 to 2320 kbps, but it's probably naive to expect it to work flawlessly at every bit rate in the range in 8 kbps increments. Although this is not stated anywhere in the chip or software documentation, it appears that its optimal bit rates are 144, 272, 400, 528, 784, 1168, 1552 and 2320 kbps. A good example is Larscom's Copper Mountain-compatible CupreDSU, which used the above 8 data rates instead of CM ones when switched into the campus modem mode in which interoperability is not a concern.

Before RS8973 appeared on the scene the early SDSL pioneers had to play with Bt8970 or even Bt8960 bitpumps and dance around the data rates those chips were designed for. Multirate capability can be implemented with external clock generation and switching circuitry, but the rates were picked through numerological games so that the analog circuitry performs well enough. This pattern is clearly visible in the Copper Mountain SDSL data rates.

There is an important test that characterises the performance of the hybrid and echo canceller at different data rates. It is called ERLE (Echo Return Loss Enhancement) and can be performed on the system with the bitpump through the BT software. The ZipWire Software User Guide explains the theory behind the ERLE test and gives instructions for performing the test and analysing the results.

All data rates ever used for SDSL

Even though the world of SDSL/2B1Q is pretty much a wild zoo, it appears that the complete list of data rates ever used by the superset of all existing flavors is not that long. It appears that all entries have come into this set from one of the following sources:

The following table lists all data rates aggregated from these sources:

Data rate    Origin
144 kbps     RS8973 pref
160 kbps     Bt8960, CM
192 kbps     Nokia Nx64
208 kbps     CM (half of 416 kbps)
272 kbps     RS8973 pref
288 kbps[*]  Bt8960
320 kbps     CM (160 * 2)
384 kbps     Nokia Nx64
392 kbps[*]  CM prototype (half of 784 kbps)
400 kbps     RS8973 pref
416 kbps     Bt8960, CM
528 kbps     RS8973 pref
768 kbps     Nokia Nx64
784 kbps     HDSL, CM, RS8973 pref
1040 kbps    CM
1152 kbps    Nokia Nx64
1168 kbps    HDSL, RS8973 pref
1536 kbps    Nokia Nx64
1552 kbps    RS8973 pref
1568 kbps    CM (784 * 2)
2320 kbps    HDSL, RS8973 pref

[*] These two data rates seem to have been used only in prototypes
    and not in any deployed SDSL product.

It is very interesting to observe how the data rates once picked rather arbitrarily by CM and Nokia have become the standard data rates for other flavors which do not resemble either CM or Nokia in any other way. The above table with the exclusion of the two starred prototype-only entries is exactly the table of all possible SDSL speeds found in the CPE products from DSLAM-neutral vendors like Netopia, and it is also the table of all possible SDSL speeds for Autobaud!