Simple answers:
1) Clock noise adds to data recovery error, so make it stable as possible.
2) mixer in frequency domain is a multiplier with matched filter bandwidths,
also, mixer in time domain it is a phase detector, in analog domain, it can be implemented, in digital domain it assumes optimal analog signal processed already ready to decimate into digital binary data.
now for a lecture... which may be redundant
There are dozens of baseband communication methods in both Analog and Digital domains.
Amplitude, Phase, Frequency methods in general leads to many more, each with tradeoffs for simplicity and performance.
In the digital realm, the signal is assumed to be noise free so SNR is not an issue, but in optical communication as in any long distance method, as in wireless or wired too, the signal is analog at some point and thus gathers noise and loss of signal contributes to SNR figure. SNR is directly related to BER on a logarithm scale. So on can value can often be calculated and correlated to the other and proved by Shannon's theorem.
So if you are only concerned with digital modulation ( ideal SNR ) ignore my comments, but real channels try to optimize SNR of received signal.
THis also includes SNR or jitter of the clock recovered from the data as excess phase error between Clock and Data can result in a decoding error.
Once this is understood, communication theory proves, helps how to improve reliable communication as fastest rate, unlike this reply.
Then optimize the channel filter to the bandwidth of the data and clock.
Some modulation methods contain DC and others invert phase after a number of bits called run length limit or RLL to avoid DC in the signal.
High speed data systems use this such. Increasing energy in the clock transitions improves SNR on recovery the clock at the expense of maximum bit rate ( low error in phase and freq., fast capture speed of PLL etc) This is why RZ is often used only in optical, since bandwidth is not the limiting factor, it may be jitter from optical path.
Some modulation methods improve SNR at the expense of upper bitrate, complexity, cost, power level or other factors.
There are volume of books on each subject mentioned above.
So one has to defined the requirement for SNR and the channel communication before choosing a receiver method.
To optimize on Optical distance before repeaters are required, this is very important.
In a lab just for ease of understanding one way to do it, it is less important.
THis is what I remember. THe details , you may choose research or not.
You can find volumes of information on the web for communication by any combinations of the following keywords:
matched filters, modulation (analog, digital), Shannon's Theorem, Baseband coding / conversion, CDR, PLL Clock recovery, eye pattern, BER, Nyquist ISI
etc.etc
I am sure you can learn to find them.
This is the purpose of your education is to learn "how to teach yourself".
But let me say I recall a solution I once used that uses a PLL to phase lock a stable clock to the center of data eye pattern others than lock to the center of the data edges or data transition window. Once you have a stable clock ( useful for both async and synchronous channels) you can filter the data to reduce jitter, phase shift, group delay distortion and thus chance of error from jitter ( or marginal SNR) YOu can correlate in time or frequency domain these values.
Now for an example of my quick web search...
basic reference...
**broken link removed** ...
definition>>
http://en.wikipedia.org/wiki/Nyquist_ISI_criterion
=Simulation results in general
http://www.docstoc.com/docs/4840553...ion-software-in-a-fiber-optic---rossir24-0019
> One implementation method >
http://www.docstoc.com/docs/27439340/A-Single-chip-Ultra-Wideband-Receiver-using-Silicon-Integrated
that concludes my sermon. :smile: now seize the channel