I'll try and answer this question . The term "fading" quite literally means a loss in the signal strength .. Wikipedia states that fading may vary with time, geographical position or radio frequency and is modeled as a random process. The statistical pdf used to represent the randomness of the fading coefficients are characterized using the rayleigh( for multipath propagation with mutliple scatterers and non-LOS link) or rician distribution(usually for Line of Sight Communication).

1. Actually, the shadowing causes fading in the signal because of obstruction(tall structures or mountains,etc) affecting the wave propagation. Shadowing and path loss can be thought of as being geographical fading.

2. Fast Fading and Slow fading are terms used to define the wireless channel characteristics in the time domain. For non-stationary mobile users, the wireless channel characteristics tend to change with time because of motion( users also experience a doppler shift too because of relative motion between users and the base station). Coherence time is defined as the time interval over which the channel characteristics remain constant and may be reduced by propagation factors such as dispersion, scattering, and diffraction. If the coherence time (Tc)is greater than symbol time(Ts)i.e Tc>Ts , then channel is said to be slow fading and fast fading otherwise

3. Frequency selective and Flat fading terms are associated with the fading in the spectral domain. A term known as coherence bandwidth is used to define range of frequencies over which the channel can be considered "flat" or constant. This is directly related to the path delay profile of the channel. When signal bandwidth exceeds the coherence bandwidth , the signal experiences frequency selective fading, i.e different frequencies experience difference levels of fading.

Hope this helps ..

Have you looked at the wiki article on fading? http://en.wikipedia.org/wiki/Fading

I'll try and answer this question . The term "fading" quite literally means a loss in the signal strength .. Wikipedia states that fading may vary with time, geographical position or radio frequency and is modeled as a random process. The statistical pdf used to represent the randomness of the fading coefficients are characterized using the rayleigh( for multipath propagation with mutliple scatterers and non-LOS link) or rician distribution(usually for Line of Sight Communication).

1. Actually, the shadowing causes fading in the signal because of obstruction(tall structures or mountains,etc) affecting the wave propagation. Shadowing and path loss can be thought of as being geographical fading.

2. Fast Fading and Slow fading are terms used to define the wireless channel characteristics in the time domain. For non-stationary mobile users, the wireless channel characteristics tend to change with time because of motion( users also experience a doppler shift too because of relative motion between users and the base station). Coherence time is defined as the time interval over which the channel characteristics remain constant and may be reduced by propagation factors such as dispersion, scattering, and diffraction. If the coherence time (Tc)is greater than symbol time(Ts)i.e Tc>Ts , then channel is said to be slow fading and fast fading otherwise

3. Frequency selective and Flat fading terms are associated with the fading in the spectral domain. A term known as coherence bandwidth is used to define range of frequencies over which the channel can be considered "flat" or constant. This is directly related to the path delay profile of the channel. When signal bandwidth exceeds the coherence bandwidth , the signal experiences frequency selective fading, i.e different frequencies experience difference levels of fading.

Hope this helps ..

Thank you all. Abhinav pinpointed for us one good point of mix-up whereby path loss is forgotten as a "fading" . Alereza, with thanks, referred to the Wikipedia page on fading (Yes, I have seen it before. Nowadays, first search results are the Wikipedia ones). Well, even in this specific page, the idea of fading is not complete as Abhinav tried to hint . The article declares fading as the "fading is deviation of the attenuation", which completely neglecting that the attenuation itself is fading as well. This may has its roots in wire-line communication in which attenuation is almost a deterministic value. For any cable its resistance per meter or its characteristic impedance can be specified and pre-designed for the given cable. In wireless, no such thing, a deterministic attenuation, exist except probably for free space in which attenuation is only distance and frequency dependent and only LOS exists. The attenuation in wireless is itself non-deterministic and described by a mean and a variance and more commonly referred to as path loss. Path loss is the result of multipath propagation as well varying with distance and geography. It is referred to in the famous fading characterizing paper of Bernard Sklar (suggested as well by Mohammad) as Large Scale Fading to characterize fading when distance changes are large (>> wave length). I like the suggestion (of Abhinav) to call it "geographical fading" or may be "Path/terrain fading instead of Path loss or attenuation.

Fading terminology still have other issues in my opinion. More suggestions are welcomed.Thanks. @AlDmour.

Rappaport's book on wireless communication presents the ideas just as explained by Abhinav. However when we model wireless channels, we usually take the fading probability distribution as normal with mean 0 and variance 1 accounting for the fact that the total power of received signal is same as transmitted one- due to signal reaching the receiver after reflections along different paths. We model the received signal as y=h*x+n

Here, y: received signal,

h: fading with CN~(0,1) as mentioned above

and n: noise which is CN~(0,N)

So, actually in modelling the channels, we use this.And this says the received signal power sums up to transmitted.

Dear Falak. Thank you for the contribution. However, allow me to say that No, it can't be that " the total power of received signal is same as transmitted one-- due to signal reaching the receiver after reflections along different paths.". Multiple paths results in signals combining constructively and destructively but they can't sum up to same signal power as for every path the signal power decays with distance. For a large number of signal paths (with no direct path) we will have a signal that fades according to Rayleigh distribution.

When we use the received signal as y=h*x+n, we have channel gain given by h. In this case h (magnitude) is described by Rayleigh fading. This means that it has a mean value (of the magnitude) lower than 1, which is fading. The zero mean of h (in your comment above : CN~(0,1) ) is for the real and imaginary parts of h (h is complex number) but the magnitude of (h) again is Rayleigh (This is the instantaneous channel gain, gain=1/loss) which has a non zero mean and a variance. Note that if h (magnitude) has zero mean then we receive 0 signal on the average (though their is a probability that h can be very small according to the distribution). Thanks. @AlDmour.

Dear Dr. Aldmour,

Thanks for the clear explanation and correction. However there's one thing towards the end that I'd like to clarify- isn't variance the signal power indicative term rather than mean? Because it contains the squared terms?

As an answer to the question by Falak above, it is important to notice that Rayleigh fading associated distribution is a simple one in which the variance and mean are not independent (e.g. like in the Gaussian distribution whereby the mean and variance can be assigned separately. The distribution fx(x)= x/σ^2. e^(-x^2/σ^2). The mean of such distribution is sqrt(π/2)*σ while the variance is 0.5*(4-π)σ^2. Thus both mean and variance of a Rayleigh parameter are proportional to σ (see for example http://mathworld.wolfram.com/RayleighDistribution.html). In other words, if the fading magnitude had a small (large) mean we shall have small(large) variance.

Thus when using Rayleigh fading in outage calculations, we normalize things by having two i.i.d CN(0,1) Gaussian variables of zero mean and unit variance. The objective shall be to determine the probability of outage when signal level drops below a certain level considering time. Therefore parameters such as level crossing rate and average fade duration that are related to speed (Doppler spread) are of importance. Notice that when considering power the distribution of fading E[h^2] becomes exponential (which lends itself for easy analysis).

I think we need now to get focused more on the original question raised by me above and request your kind contribution. Thanks. @AlDmour