Hi i am designing 4.5KW Phase shift full bridge converter. What are all precautions to be taken from EMI/EMC point of view

The input voltage is 90-210 output voltage is 135-216. the output current is 0-30A.

There are many different forms of phase shift converter. Will you be using the traditional design where the switching devices are MOSFETs and the parasitics are employed to provide resonant transitions? If so, then the output current range will mean that you will be hard-switching at low load. Hard-switching = high noise generally. All standard noise control techniques will be needed. Since both legs of the bridge will be noisy, you will need a screen in the transformer. Keep the bridge, transformer and link caps as tight a layout as possible. Maybe a slightly non-standard full bridge is needed. At a previous company I designed a 20kW ZVS/ZCS converter using IGBTs and a saturable reactor to widen the soft switching range. The IGBT's were special zero tail parts from Semikron. The saturable reactor was water cooled. If I was doing one again I would try the ZVS/ZCS converter with secondary active clamp in Cho, Lee and Rim's IEEE paper from 1996. Alternatively, since the power will be low near zero output voltage, maybe you could control the EMI by reducing the gate current to the switching devices, but this will only be effective using MOSFETS since the turn-off of IGBT's can't be altered much in this way.

I don't believe that venkata gave a switching frequency, so how it can be said that at the frequencies he's using, that an input line filter is a must, I don't know. It depends on your input EMI level and your EMI specifications-period. Either an input filter is needed or it's not. Having said that, there is a lot that can be done to mitigate needing a filter when you may be able to get away without one. Kok mentioned many things and if I'm deciphering them correctly, I think he's generally correct. But there are two points I'll emphasize here. First, good PWB layout (and don't even think about using a breadboard unless you're operating below 20KHz and even then, it would tell you little about EMI on the finished PWB circuit) is not only a must, but it will save you countless hours later on in reducing EMI and circuit noise that can make you're controller IC go crazy. At those power levels and a more likely frequency of 100KHz, ground planes are a must, and that probably means a minimum 4-layer PWB and if your currents are high, maybe 4 to 6 oz copper per layer (at least on the outer layers) may be necessary. Be very careful about the UFasts or Schottkys and not just the absolute value of the Trr, vut also the shape. Most desireable is a low Trr with a soft (not-too-abrupt) switching characteristic. I know this can be hard to find as many semi mfgs don't supply much of a datasheet. I'f EMI is a critical factor in your converter, you probably do need a RC or RCD snubber on the output rectifier or rectifiers. Sometimes, as Kok said, a ferrite bead can do the job better and with less loss. This may partially be a matter of experimentation using a spectrum analyzer with an LISN or current probe (for conducted EMI), or antenna's and a shielded area for radiated EMI.

I suppose you will use the most common widely used phase shift FB topology in your design. Highest du/dt will be appeared when main SW is turned on at light load. Highest di/dt will be appeared when main SW is turned off at full load. These two parameters will rely on circuit parasitic parameters and resonant component values in the tank. They will determine the noise level of converter on SW side. Reverse recovery noises on rectifier side also rely on all these parameters.

Conductive EMI (differential mode and common mode) can be well predicted thru simulation if the converter, EMI filter and EMI LISN circuits are built in right way. That will be helpful for EMI precautions.

I totally agree with Bill's comment about being able to predict EMI through simulation. Many people believe that EMC is a black art. All the biggest power electronics problems are at low frequency in my experience and the circuit behavior can be modelled with good accuracy. The key to getting a good model is good measurement of all the circuit parts and parasitics. A wide bandwidth network analyser is essential.

In order to evaluate the suggestions listed above, it would help to operate the power circuit in open loop mode (i.e. constant duty cycle) at rated power (or higher). Since the control loop is taken out of the equation, any circuit misbehavior can be definitely attributed to EMI. One can then optimally select gate drive resistors, R-C snubbers and diodes of the appropriate quality to meet the requisite EMI standards.

I can't say simulation no help in design or any technical study or any advance technology advancement, we need simulation. Is agreeable simulation shall help but it only valuable when come to you have the "sense" - what to "input" and know what to look at.

Not able to get re-presentable modelling of the component and get re-presentable switching mechanism, the result is just for references only !

A powerful simulation software that can merge layout, mechanical characteristic, electrical circuit simulation, and not to miss have an individual or a group of strong technical expertise that help to get the power train component modeled, miss one of it will cause Actual and Simulation has a "gap".

Please do not use correction here and there to say your simulation is very close to actual, in fact you have alter actual component characteristic or adding none re-presentable parameters, and try to match with actual, adjusting the correction probably can help to match for 1 condition but when condition change then try to match again, this should not be the way we use simulation.

Conducted EMI, Low frequency and some high frequency is roughly can be estimated, if know the switching waveform and knowing what critical parameter of critical component, without simulation probably almost know how sever the EMI will be, but not quantitatively. You can get some of the literature that talk about Fourier transformation, you will find it is useful for how to predict the noise.