Few analogies proving the subject:-

• If the distance is to short and the attenuator is too close to the transmitter, the reflected light off the attenuator will be directed back towards the Tx laser. Which will also blow your transmitter.so we place it at Rx.

• Also keeping attenuator at Rx will attenuate the noise along with the signal.

• The most important reason for putting them on the RX side is that you are protecting that which needs to be protected - the receiver in your optics. This way you know that you're not going to potentially blow the receiver in your optics by plugging in too large a signal because you assumed there was an attenuator on the TX at the far end, and there wasn't.

• It's more convenient to test the receiver power before and after attenuation or while adjusting it with your power meter at the receiver, plus any reflectance will be attenuated on its path back to the source.

Keynote on Using Attenuators With Fiber Optic Data Links

The ability of any fiber optic system to transmit data ultimately depends on the optical power at the receiver as shown above, which shows the data link bit error rate as a function of optical power at the receiver. (BER is the inverse of signal-to-noise ratio, e.g. high BER means poor signal to noise ratio.)  Either too little or too much power will cause high bit error rates. 

Too much power, and the receiver amplifier saturates, too little and noise becomes a problem as it interferes with the signal. This receiver power depends on two basic factors: how much power is launched into the fiber by the transmitter and how much is lost by attenuation in the optical fiber cable plant that connects the transmitter and receiver.

If the power is too high as it often is in short singlemode systems with laser transmitters, you can reduce receiver power with an attenuator. Attenuators can be made by introducing an end gap between two fibers (gap loss), angular or lateral misalignment, poor fusion splicing (deliberately), inserting a neutral density filter or even stressing the fiber (usually by a serpentine holder or a mandrel wrap). Attenuators are available in models with variable attenuation or with fixed values from a few dB to 20 dB or more.

 
Gap-loss attenuators for multimode fiber
 

Serpentine attenuators for singlemode fiber
 

Generally, multimode systems do not need attenuators. Multimode sources, even VCSELs, rarely have enough power output to saturate receivers. Singlemode systems, especially short links, often have too much power and need attenuators.

For a singlemode applications, especially analog CATV systems, the most important specification, after the correct loss value, is return loss or reflectance! Many types of attenuators (especially gap loss types) suffer from high reflectance, so they can adversely affect transmitters just like highly reflective connectors. 

Choose a type of attenuator with good reflectance specifications and always install the attenuator ( X in the drawing) as shown at the receiver end of the link. This is because it's more convenient to test the receiver power before and after attenuation or while adjusting it with your power meter at the receiver, plus any reflectance will be attenuated on its path back to the source. 

Test the system power with the transmitter turned on and the attenuator installed at the receiver using a fiber optic power meter set to the system operating wavelength. Check to see the power is within the specified range for the receiver.

If the appropriate attenuator is not available, simply coil some patchcord around a pencil while measuring power with your fiber optic power meter, adding turns until the power is in the right range. Tape the coil and your system should work. This type of attenuator has no reflectance and is very low cost! The fiber/cable manufacturers may worry about the relaibility of a cable subjected to such a small bend radius. You should probably replace it with another type of attenuator at some point, however.

Singlemode attenuator made by wrapping fiber or simplex cable around a small mandrel. This will not work well with bend-insensitive fiber.

ref:http://www.thefoa.org/tech/ref/appln/attenuators.html

Non-linear interactions between the signal and the silica fibre transmission medium begin to appear as optical signal powers are increased to achieve longer span lengths at high bit rates. Consequently, non-linear fibre behaviour has emerged as an important consideration both in high capacity systems and in long unregenerated routes. These non-linearities can be generally categorized as either scattering effects (stimulated Brillouin scattering and stimulated Raman scattering) or effects related to the fibre's intensity dependent index of refraction (self-phase modulation, cross-phase modulation, modulation instability, soliton formation and four-wave mixing). A variety of parameters influence the severity of these non-linear effects, including line code (modulation format), transmission rate, fibre dispersion characteristics, the effective area and non-linear refractive index of the fibre, the number and spacing of channels in multiple channel systems, overall unregenerated system length, as well as signal intensity and source line-width. Since the implementation of transmission systems with higher bit rates than 10 Gbit/s and alternative line codes (modulation formats) than NRZ-ASK or RZ-ASK, described in [b-ITU-T G-Sup.39], non‑linear fibre effects previously not considered can have a significant influence, e.g., intra‑channel cross-phase modulation (IXPM), intra-channel four-wave mixing (IFWM) and non‑linear phase noise (NPN).

Refer ITU-T G.663 for further study.

General Causes of Bit Errors

 Performance degrade of key boards
Abnormal optical power
Signal-to-noise ratio decrease
Non-linear factor
Dispersion (chromatic dispersion/PMD) factor
Optical reflection
External factors (fiber, fiber jumper, power supply, environment and others)

The main advantages and drawbacks of EDFAs are as follows.

Advantages

• Commercially available in C band (1,530 to 1,565 nm) and L band (1,560 to 1,605) and up to  84-nm range at the laboratory stage.
• Excellent coupling: The amplifier medium is an SM fiber;
• Insensitivity to light polarization state;
• Low sensitivity to temperature;
• High gain: > 30 dB with gain flatness < ±0.8 dB and < ±0.5 dB in C and L band, respectively, in the scientific literature and in the manufacturer documentation
• Low noise figure: 4.5 to 6 dB
• No distortion at high bit rates;
• Simultaneous amplification of wavelength division multiplexed signals;
• Immunity to crosstalk among wavelength multiplexed channels (to a large extent)

Drawbacks

• Pump laser necessary;
• Difficult to integrate with other components;
• Need to use a gain equalizer for multistage amplification;
• Dropping channels can give rise to errors in surviving channels:dynamic control of amplifiers is  necessary.

The ITU standards define a “suspect internal flag” which should indicate if the data contained within a register is ‘suspect’ (conditions defined in Q.822). This is more frequently referred to as the IDF (Invalid Data Flag).

PM is bounded by strict data collection  rules as defined in standards. When the collection of PM parameters is affected then  PM system labels the collection of data as suspect with an Invalid Data Flag (IDF). For the sake of identification; some unique flag  is shown next to corresponding counter.

The purpose of the flag is to indicate when the data in the PM bin may not be complete or may have been affected such that the data is not completely reliable. The IDF does not mean the software is contingent.

Some of the common reasons  for setting the IDF include:

• a collection time period that does not start within +/- 1 second of the nominal collection window start time.
• a time interval that is inaccurate by +/- 10 seconds (or more)
• the current time period changes by +/- 10 seconds (or more)
• a restart (System Controller restarts will wipe out all history data and cause time fluctuations at line/client module;  a module restart will wipe out the current counts)
• a PM bin is cleared manually
• a hardware failure prevents PM from properly collecting a full period of PM data (PM clock failure)
• a protection switch has caused a change of payload on a protection channel.
• a payload reconfiguration has occurred (similar to above but not restricted to protection switches).
• an System Controller archive failure has occurred, preventing history data from being collected from the line/client  cards
• protection mode is switched from non-revertive to revertive (affects PSD only)
• a protection switch clear indication is received when no raise was indicated
• laser device failure (affects physical PMs)
• loss of signal (affects receive - OPRx, IQ - physical PMs only)
• Control Plane is booted less than 15 min period for 15-min interval and less than 24 hour period for 24-hour interval.

Suspect interval is determined by comparing nSamples to nTotalSamples on a counter PM. If nSamples is not equal to nTotalSamples then this period can be marked as suspect. 

If any 15 minute is marked as suspect or reporting for that day interval is not started at midnight then it should flag that 24 Hr as suspect.

Some of the common examples are:

• Interface type is changed to another compatible interface (10G SR interface replaced by 10G DWDM interface),
• Line type is changed from SONET to SDH,
• Equipment failures are detected and those failures inhibit the accumulation of PM.
• Transitions to/from the 'locked' state.
• The System shall mark a given accumulation period invalid when the facility object is created or deleted during the interval.
• Node time is changed.

Why does FEC introduce latency?

The addition of FEC overhead at the transmit end and FEC decoding at the receive end cause latency in signal transmission.

Relationship between latency and FEC scheme types

• To improve the correction capability, more powerful and complex FEC codes must be used. However, the more complex the FEC codes are, the more time FEC decoding will take. As a result, the latency is increased.
• For a non-coherent 40G signal, AFEC typically introduces a larger latency (60~120 us) than FEC, which introduces a latency of 30 us.
• Latency of a non-coherent 40G signal is related to the ODUk mapping mode.
• Latency is also related to the amount of overhead. More overhead means that FEC decoding will take more time.
• Latency introduced by 100G SD-FEC with 20% overhead > Latency introduced by 100G HD-FEC with 7% overhead
• In addition, the latency is subject to the signal coding rate. With the overhead unchanged, there will be less latency as the signal rate increases.
• Latency introduced by 100G HD-FEC < Latency introduced by 40G AFEC

Latency specifications of OTN equipment

Data and storage services are sensitive to latency while OTN services are not. Currently, no international standards have defined how much latency that OTN signals must satisfy. Vendor equipment supports latency testing for different service rates, FEC schemes, and mapping modes. You can recommend a latency value for customers, but latency should not be an acceptance test item. You cannot make a commitment to a latency specification.

In a non-coherent WDM system, each optical channel on the line side uses only one binary channel to carry service information. The service transmission rate on each optical channel is called bit rate while the binary channel rate is called baud rate. In this sense, the baud rate was equal to the bit rate. The spectral width of an optical signal is determined by the baud rate. Specifically, the spectral width is linearly proportional to the baud rate, which means a higher baud rate generates a larger spectral width.

• Baud (pronounced as /bɔ:d/ and abbreviated as "Bd") is the unit for representing the data communication speed. It indicates the signal changes occurring in every second on a device, for example, a modulator-demodulator (modem). During encoding, one baud (namely, the signal change) actually represents two or more bits. In the current high-speed modulation techniques, each change in a carrier can transmit multiple bits, which makes the baud rate different from the transmission speed.

In practice, the spectral width of the optical signal cannot be larger than the frequency spacing between WDM channels; otherwise, the optical spectrums of the neighboring WDM channels will overlap, causing interference among data streams on different WDM channels and thus generating bit errors and a system penalty.

For example, the spectral width of a 100G BPSK/DPSK signal is about 50 GHz, which means a common 40G BPSK/DPSK modulator is not suitable for a 50 GHz channel spaced 100G system because it will cause a high crosstalk penalty. When the baud rate reaches 100 Gbaud/s, the spectral width of the BPSK/DPSK signal is greater than 50 GHz. Thus, it is impossible to achieve 50 GHz channel spacing in a 100G BPSK/DPSK transmission system.

(This is one reason that BPSK cannot be used in a 100G coherent system. The other reason is that high-speed ADC devices are costly.)

A 100G coherent system must employ new technology. The system must employ more advanced multiplexing technologies so that an optical channel contains multiple binary channels. This reduces the baud rate while keeping the line bit rate unchanged, ensuring that the spectral width is less than 50 GHz even after the line rate is increased to 100 Gbit/s. These multiplexing technologies include quadrature phase shift keying (QPSK) modulation and polarization division multiplexing (PDM).

As we know that either homodyne or heterodyne detection can be used to convert the received optical signal into an electrical form. In the case of homodyne detection, the optical signal is demodulated directly to the baseband. Although simple in concept, homodyne detection is difficult to implement in practice, as it requires a local oscillator whose frequency matches the carrier frequency exactly and whose  phase is locked to the incoming signal. Such a demodulation scheme is called synchronous and is essential for homodyne detection. Although optical phase-locked loops have been developed for this purpose, their use is complicated in practice.

Heterodyne detection simplifies the receiver design, as neither optical phase locking nor frequency matching of the local oscillator is required. However, the electrical signal  oscillates rapidly at microwave frequencies and must be demodulated from the IF bandto the baseband using techniques similar to those developed for microwave communication systems. Demodulation can be carried out either synchronously or asynchronously. Asynchronous demodulation is also called incoherent in the radio communication literature. In the optical communication literature, the term coherent detection is used in a wider sense.

A lightwave system is called coherent as long as it uses a local oscillator irrespective of the demodulation technique used to convert the IF signal to baseband frequencies.

*In case of homodyne coherent-detection technique, the local-oscillator frequency is selected to coincide with the signal-carrier frequency.

*In case of heterodyne detection the local-oscillator frequency  is chosen to differ from the signal-carrier frequency.