SD cable eq issues with LMH0387

Hi Carl,

I too have the same issues as you.  Did you manage to resolve them?

Cheers,

Richard

Richard, Carl

Could you describe the setup that you are using? - is this on one of our own evaluation boards, or is it on your design, and if your design, do you have access to the SPI bus?

Hi Mark,

Yes this is our own board, and yes we have access to the SPI bus.

Cheers,

Richard

Richard

A few questions:   If you use the SPI bus, and read registers 2 and 3 - register 3 is the cable length indication, and it will tell us how long of a cable the device thinks that it is seeing - the larger the value the longer the cable that the part thinks that it is seeing - with a cable length of about 100m, I would expect to see a reading of 10-15, if it is at the max (25) that would indicate that the equalization control is cranked all the way up.

Register 2 is an adjustment to DC gain at the front end of the equalizer.  Assuming a normal setup, this should be 00110000.

Also, just as a sanity check - the LMH0387 is internally terminated, if there is an additional external termination, this will attenuate the signal, and the cable reach will be smaller, and the device is optimized for Belden 1694A cable - if you are using another cable, performance will be different from what is in the datasheet.

Mark

Hi Mark,

The cable length indication seems to be about right.  And yes we've set up Reg 2 correctly.

When you say "the LMH0387 is internally terminated" are you referring to the SDI BNC connection? Or the connection to the fpga?  If it's the former then we don't have any additional termination on there.

One thing we have noticed is that pin 17 (TermRx) is very sensitve.  If we place a scope probe on this pin then lots of errors get introduced.  Is that what you'd expect?

Cheers,

Richard

Richard,

Regarding the comment on the LMH0387 "internal termination," we were referring to the BNC_IO pin.  The 75Ω termination and return loss network are internal, so the only external component should be the AC-coupling capacitor.

The TERM_RX termination is required to balance the input network to the equalizer portion of the LMH0387.  This termination should be kept short and treated as 75Ω, similarly to the BNC_IO trace.  It is a path to the equalizer input, so noise on this pin could affect equalizer performance.

Can you clarify the cable reach you are observing with the LMH0387?  Is the issue only at SD, or at HD and 3G rates as well?  Have you compared the performance using one of our LMH0387 EVMs?

Also note that it is important to isolate the high-gain equalizer input from noise and potential crosstalk.  Are you aware of nearby noise or signals that could be affecting the equalizer performance?

Gary Melchior

Hello Mark,

I'm experiencing similar problems as mentioned above and I receive very poor quality from the equalizer.

When working with SD (270Mbps) and applying 12dB attenuation on the SDI input, I see that CLI is almost 25. Another step to 13dB attenuation and the EQ starts to output garbage. I expect that CLI will be around 10 in this configuration.

12dB attenuation is equal to around 100m cable in SD, whereas CLI reports that cable length is ~350m.

It looks like the EQ is getting to its max gain too early.

Our design is based on the datasheet. Without attenuation I measure around 700mV swing after the input AC coupling capacitor.

Changing the launch amplitude configuration helped a bit, but it's still far from spec.

Is there any point to try to change TERMRX components? 

I'd appreciate your help.

BTW, in case of 700mV input swing, how do you suggest to configure the launch amplitude register?

Thanks,

Amir

How are you applying the attenuation?  The cable equalizers anticipate that the frequency characteristic of the path between the driver and the cable equalizer will be cable which has losses dominated by the skin effect.  The cable equalizer is not designed to equalize for fixed attenuations, and will not operate correctly under these conditions.   Does the equalizer behave properly when equalizing for cable interconnect?

Hi Mark,

I use RF 75ohm attenuator designed for these frequencies.

When I work with just standard RG6 long cable (around 40m) I get a value of around 15 in the CLI register (~150m according to CLI graph)

In addition, when I use other devices (blackmagic SDI card, SDI to HDMI converter, external displays and other device we have based on Gennum SDI reciever) I see that they behave much better with the attenuator and the long cable. 

Just for the comparison, when using LMH0387 we can work up to 12dB attenuation whereas Blackmagic and Gennum receiver work well up to 18dB and 26dB respectively.

When I configure register 0x2 to 0xB8, meaning "Coarse Control" enabled and fine tune is set to -2% we get the best results (around 16-18dB).

Is there any down side to work in this configuration? Can we have too much gain in case we work with short cable?

I'd appreciate your support on this matter.

Thanks,

Amir

Amir

The automatic equalization always applies a high pass filter whcih has the inverse characteristic of cable.  Different cables will have different amounts of loss per unit length, but they all have a similar shape where the attenuation is proportional to SQRT(Freq).   If you need to compensate for a fixed attenuator, then you can use the coarse control bit of register 02, which will compensate for a fixed 6dB attenuation.   Small differences that might be expected from slightly large or small launch amplitudes will be corrected for automatically.

If you set the coarse control register and do not have a fixed atteuator in your path, you will have poor performance at some cable lengths.

The graph in the datasheet for Cable Length vs CLI is for using Belden 1694A cable, with no additional attenuators.  If you use a different cable, you will need to generate a new curve.