Forum Discussion

Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
16 years ago

Stratix II GX Transceiver: Does 8b/10b coding work properly?

Hi again,

I have limited the problem of undefined bits in the tx-output datastream to a problem with the 8b/10b-encoder.

In ModelSim, I found the internal signal between the Phase Compensation FIFO Buffer and the 8b/10b encoder. The dwords from the fifo are ok.

The output wire of the 8b/10b encoder contains some undefined bits I have no explanation for. Here is the signal example:

FIFO-in: ... FF0F33FB.....................................................................................

FIFO-out: ...| FF0F | 33FB | FF0F | 33FB | ....

8b/10-out: ...|1001010011U00U011011 | U00U1101010010111010 | 1001010011U00U011011 | U00U1101010010111010 | ...

I expected to see the follwowing, 8b/10b- encoded words:

1010110001 (RD+) or 0101001110 (RD-) for FF

0101110100 1010001011 for 0F

1100101001 1100101001 for 33

1101100001 0010011110 for FB

Of course, the serialized data is wrong, too. I have no idea how to fix that problem and really need some help..

daniel

4 Replies

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    16 years ago

    Problem solved! I assumed that if the tx_ctrlenable port is not used, all words would be handled as data (Dx.y). That was wrong! At every point in my datastream including a word which also is part of the control-words (Kx.y) which are listed in the table 6-2 of the stratix II gx-handbook (vol. 2)), I got undefined output of 8b/10b encoding. Now I asserted the tx_ctrlenable to "0000" and it works fine.

    The next problem is that word alignment on the receiver-side does not work. after a reset I do a rx_enapatternalign (rising edge) whith an alignment pattern (x"0000BC1C") in my input data (20b after 8b10b-endoding, the word to specify in the MegaWizard?). tx_ctrlenable is set to "1111", so I expect to get both the rx_patterndetect and rx_syncstatus high. That is not the case and I don't know how to fix. The rx_dataout is always 9C9C9C9C after the reset time.

    I hope somebody sees whats wrong here. Thank you for an answer,

    daniel
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    16 years ago

    Hi Daniel,

    I have a question for you which is a bit off topic, but anyway.

    I also have a Stratix II GX kit with an SFP module for optical transmission and would like to implement a loop-back test that you are implementing now.

    The problem is that my SFP module uses NRZ encoding to transmit the data. I checked the SFP specification and actually it states that the modules can use 8b10, 4b5b, NRZ, Manchester or "Unspecified" coding. Since the SFP module is encoding the data i cannot use the 8b10b decoder in the ALTGX megafunction: i need to decode it using NRZ first and then apply a 8b10b decoder.

    Have you come across this problem? Any idea how to solve it (besides building the PCS layer myself)?

    Faruk
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    16 years ago

    Hi Faruk,

    I'm sorry, but I have not much experience with SFP modules. But my suggestion is to try the "unspecific" mode in combination with the 8b/10b encoding of the fpga. In my case, there is a possibility to configure the SFP-Module via I2C bus. Fortunetely, the default mode is "unspecific" and I don't need to use this.

    With using NRZ, I am afraid you will get problems with clock recovery, too. As I know, you need a seperate clock signal for NRZ and everything could become complicated... Is there a special reason for using NRZ?

    In between, I solved my problem with 8b/10b coding. I didn't notice that every byte is transferred with the least significant bit first. So, in the MegaWizard, you have to write down the 8b/10b code (20b) of the LSB-first-ordered bytes of the pattern bytes (usually Comma, BC). Additionally, while sending this the tx_ctrlenable has to be "0011", indicating that the two right bytes 0000BC1C are control bytes (K28.5 and K28.0). having those signals set, a single rx_enapatternalign on a rising edge works and the rx_syncstatus gets high (1111). with this signal, it is possible to set tx_ctrlenable to 0000 again.

    Daniel
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    Hi Daniel,

    I'm working in the same topic as you encoding with 8b10b using SIIGX. I'm trying to syncrhonize my transceivers but I don't achieve it.

    Could you send me any example code??

    Thanks, Fer.