My work starts with a supposition of $N$, so that for $n > N$ we have $\vert b \vert ^n < \epsilon$.
Since $0 < \vert b \vert < 1$, we see the logarithm with base $\vert b \vert$ is a decrescent function meaning it will invert the inequality once taken. $$\vert b \vert ^n < \epsilon $$ $$n > \text{log}_{\vert b \vert}\epsilon $$
Done the scrapwork, lets start the formal proof. Let $\epsilon > 0$. Let $N = \text{log}_{\vert b \vert}\epsilon$. If $n > N$ and $0 < \vert b \vert< 1$, then $$\vert b \vert ^n < \vert b \vert ^{\text{log}_{\vert b \vert}\epsilon} = \epsilon$$ Hence by definition, $\text{lim} \ b^n=0$
Have I done everything correctly? I am using in an implicit manner that $\vert b^n \vert = \vert b \vert ^n$, is everything fine with this? (Something is really pinching me up!)
You can make janmarqz's result more explicit.
Since $|b^n| = |b|^n$, we can assume that $0 < b < 1$.
Then $b = \dfrac1{1+a}$ with $a > 0$, so that, from Bernoulli's inequality, $(1+a)^n \ge 1+an $.
Therefore, since $a = \dfrac1{b}-1$,
$\begin{array}\\ b^n &=\dfrac1{(1+a)^n}\\ &\le \dfrac1{1+na}\\ &\lt \dfrac1{na}\\ &=\dfrac1{n(\frac1{b}-1)}\\ \end{array} $
To make $b^n < \epsilon$, it is enough if $\epsilon > \dfrac1{n(\frac1{b}-1)}$ or $n > \dfrac1{\epsilon(\frac1{b}-1)}$.
This is, of course, far worse than using the log, but it is completely elementary.
This is also not original with me. I first saw this in "What is Mathematics?" by Courant and Robbins, which I highly recommend and is available in paperback for less than $20.